Background:Stress fractures (SFs) are one of the more severe overuse injuries in military training, and therefore, knowledge of potential risk factors is needed to assist in developing mitigating strategies.Purpose:To develop a prediction model for risk of SF in Royal Marines (RM) recruits during an arduous military training program.Study Design:Case-control study; Level of evidence, 3.Methods:RM recruits (N = 1082; age range, 16-33 years) who enrolled between September 2009 and July 2010 were prospectively followed through the 32-week RM training program. SF diagnosis was confirmed from a positive radiograph or magnetic resonance imaging scan. Potential risk factors assessed at week 1 included recruit characteristics, anthropometric assessment, dietary supplement use, lifestyle habits, fitness assessment, blood samples, 25(OH)D, bone strength as measured by heel broadband ultrasound attention, history of physical activity, and previous and current food intake. A logistic least absolute shrinkage and selection operator (LASSO) regression with 10-fold cross-validation was used to select potential predictors among 47 candidate variables. Model performance was assessed using measures of discrimination (c-index) and calibration. Bootstrapping was used for internal validation of the developed model and to quantify optimism.Results:A total of 86 (8%) volunteer recruits presented at least 1 SF during training. Twelve variables were identified as the most important risk factors of SF. Variables strongly associated with SF were age, body weight, pretraining weightbearing exercise, pretraining cycling, and childhood intake of milk and milk products. The c-index for the prediction model, which represents the model performance in future volunteers, was 0.73 (optimism-corrected c-index, 0.68). Although 25(OH)D and VO₂max had only a borderline statistically significant association with SF, the inclusion of these factors improved the performance of the model.Conclusion:These findings will assist in identifying recruits at greater risk of SF during training and will support interventions to mitigate this injury risk. However, external validation of the model is still required.