Work

Game-Level Temporal Patterns Identify Optimal Preinjury Windows for Shoulder and Elbow Injuries in Major League Baseball Pitchers

Year: 2026

Type: article

Abstract: Background: Shoulder and elbow injuries are a major concern for professional baseball pitchers, highlighting the need for prevention. Prior studies in Major League Baseball (MLB) rely on season-level metrics, potentially masking short-term changes preceding injury. Purpose: To identify discrete, optimal preinjury windows for predicting shoulder and elbow injuries in MLB pitchers using machine-learning models applied to game-level workload and mechanical data. Study Design: Case-control study; Level of evidence, 3. Methods: Game-level MLB pitcher data (2017-2025) were obtained from publicly available databases. Shoulder and elbow injuries were identified using injured list records. Gradient-boosted decision tree models (XGBoost) were trained to distinguish pitcher-games occurring within predefined preinjury windows (7-49 days) versus control observations. Optimal windows were identified using changepoint analysis of cross-validated area under the curve (AUC). Feature importance characterized baseline risk, and dynamic analyses assessed short-term variability. Models were stratified by injury type and pitching role. Results: Cohorts included 3970 pitcher-seasons (116,260 pitcher-games) for elbow injuries and 4085 pitcher-seasons (120,251 pitcher-games) for shoulder injuries. Elbow injuries occurred in 430 pitcher-seasons and shoulder injuries in 545 pitcher-seasons. Optimal preinjury windows were 27.5 to 29.5 days for starters and 19.7 to 21.3 days for relievers for elbow injuries, and 22.8 to 25.7 days for starters and 18.6 to 20.3 days for relievers for shoulder injuries. Model discrimination was strong (elbow, AUC 0.815-0.868; shoulder, AUC 0.797-0.815). Feature importance identified the per-game metrics most influential for model injury discrimination: release position, pitch count, slider proportion, and velocity variability (elbow), and release position, spin rate, maximum velocity, and age (shoulder). Dynamic game-to-game fluctuations in pitch count consistently contributed to risk for both elbow (+1.6%-6.2%) and shoulder (+3.3%-10.4%) injuries. Conclusion: Using routinely collected game-level pitching data, a machine-learning model identified discrete, optimal preinjury windows for predicting elevated shoulder and elbow injury risk in MLB pitchers: ~4 weeks for starters and ~3 weeks for relievers. Baseline per-game metrics defined underlying injury risk, whereas short-term game-to-game variability contributed to dynamic increases in modeled risk. These findings suggest that injury risk may be detectable several weeks before injured-list placement, providing a potential window for closer monitoring and workload evaluation.

Source: Orthopaedic Journal of Sports Medicine

DOI: 10.1177/23259671261477532