University of Patras
Επιδημιολογική ανάλυση και μοντελοποίηση των παραγόντων κινδύνου τραυματισμών του ώμου σε αθλούμενους του CrossFit στην Ελλάδα
Abstract
dc:descriptionBackground: The shoulder joint is the most frequently injured area in CrossFit (CF) with significant functional and performance consequences. The multifactorial aetiology of these injuries limits the effectiveness of conventional linear statistical approaches in identifying prognostic indicators. Objective: This doctoral thesis investigates the epidemiology of shoulder injuries among CF athletes in Greece. It aims to determine risk factors and develop a predictive model using advanced Machine Learning (ML) methods. Methods: The investigation comprised three main stages. First, a sport-specific field test – the CrossFit Functional Assessment Battery for Shoulder joint (CF-FABS) – was developed and its inter-rater and test-retest reliability were examined. Second, a 12-month prospective cohort of 109 CF athletes was established to monitor incident shoulder injuries. Third, demographic, epidemiological, and functional data were integrated to profile athletes; risk factors were explored using linear and non-linear analyses (logistic regression, PCA, FAMD); and a predictive model was built to estimate future shoulder injury. A Random Forest model was trained with 100k-fold cross-validation and interpreted using SHAP values. Results: The inter-rater reliability analysis showed a Cohen’s κ > 0.82 across all CF-FABS individual tests, with percentage agreement ≥ 90%, while the test-retest reliability showed κ > 0.66 with agreement ≥ 75%. The Injury Incidence Rate (IIR) was calculated as 0.79 injuries per 1,000 hours of exposure. Most injuries involved the rotator cuff (45%) and were primarily classified as overuse injuries (80%). Shoulder injuries were most prevalent during Olympic weightlifting exercises (45%). Linear statistical analyses and modelling did not show any statistically significant correlations with injury occurrence. Advanced ML identified a best-performing Random Forest model with a mean accuracy of 63%, indicating low overfitting. The model revealed the critical parameters of: side-to-side ROM deficits in internal rotation, external rotation, and flexion; prior injury to the non-dominant and dominant shoulder; pain during CF participation; and bilateral ROM symmetries in the aforementioned movements. Conclusion: CF-FABS is the first reliable, sport-specific, field-based tool to facilitate the detection of biomechanical deficits by coaches and researchers. Prior shoulder injury, pain during CF, and side-to-side differences in ROM constitute salient features for injury occurrence. Compared to traditional approaches, ML provides an interpretable and practical predictive framework capable of supporting early identification of higher-risk athletes and the design of targeted prevention strategies.
Degree
thesis:*- Grantor dc:publisher
- University of Patras
- Year dc:date
- 2025
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Bakaraki, Akrivi
- Μπακαράκη, Ακριβή
Subjects
dc:subject × 15Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/60378