Beyond the conventional: Artificial intelligence in identifying risk factors in sports injuries. A scoping review.

Clinical Snapshot

Key Findings

• P Value: Not reported at review level

• Effect Size: Not applicable — scoping review with narrative synthesis; no pooled effect size calculated

• Primary Outcome: Mapping of AI methods used to identify conventional and unconventional risk factors for sports injuries across 59 included studies

• Nnt Or Sensitivity: Classification performance described as 'acceptable but variable' across included studies; no aggregated sensitivity, specificity, AUC, or NNT reported. Specific metrics from individual studies are not summarised in the abstract.

• Confidence Interval: Not reported — narrative synthesis design precludes confidence interval estimation

Clinical Application

Clinical implementation of AI-based injury risk tools remains premature based on this review's findings. Key barriers include lack of external validation, heterogeneous measurement standards, absence of standardised reporting, and limited evidence of clinical integration pathways. The review identifies the need for multicenter prospective cohorts and transparent model sharing before routine clinical use. Cost, infrastructure requirements for multimodal sensor data collection, and clinician training are additional implementation barriers not addressed in the review. In Australia, sports injury prevention is a priority area for organisations including Sports Medicine Australia, the Australian Institute of Sport (AIS), and state-based sporting bodies. The RACGP does not currently have specific guidelines on AI-assisted injury risk assessment. No AI-based sports injury risk tool is currently listed on the TGA's Australian Register of Therapeutic Goods (ARTG) as a regulated Software as a Medical Device (SaMD), though this regulatory pathway exists under TGA's Digital Health framework. PBS listing is not applicable to AI diagnostic or risk assessment tools. Australian sports medicine practitioners should note that the evidence base identified in this review is insufficient to support adoption of AI injury risk tools in routine clinical or high-performance sport settings without further validation in Australian athlete populations. The AIS's ongoing investment in athlete monitoring technology makes this an area of active relevance, but clinical governance frameworks for AI tool adoption remain underdeveloped nationally. Athletes across competitive and recreational sports disciplines where injury risk monitoring is clinically relevant. Most applicable to sports medicine physicians, physiotherapists, and high-performance sport scientists working with elite or semi-elite athletes. Community-level applicability is uncertain given the technology requirements of multimodal sensor-based AI approaches.

Abstract

PURPOSE: To map contemporary uses of artificial intelligence (AI) to identify conventional and unconventional risk factors for sports injuries in athletes. METHODS: Following Joanna Briggs Institute guidance and PRISMA-ScR, we conducted an open-access search (January 2015-September 2025) in PubMed/MEDLINE, Scopus, and ScienceDirect using MeSH/DeCS terms for AI and injury risk. Records were managed in Rayyan with de-duplication, independent triple screening, standardized data extraction, and narrative synthesis to address heterogeneity. RESULTS: Fifty-nine studies met inclusion criteria. Frequently modelled predictors included sleep quality and psychological state, external and internal load metrics, and prior injury. AI approaches spanned classical machine learning and deep learning applied to multimodal sensor and clinical data. Across studies, generalizability was limited by heterogeneous populations, outcomes, and measurement/reporting practices; few works reported consistent measurement standards or external validation. Where assessed, observer agreement and classification performance were acceptable but variable. CONCLUSION: AI for sports-injury risk is expanding rapidly, led by classical machine learning on multimodal sensor data. Key gaps are external validity and reproducibility. Progress will require multicenter prospective cohorts, standardized measurement and reporting, sport-specific meta-analyses, transparent model sharing, and deliberate clinical integration into athlete care pathways.

References

1. 1.Martinez Mora, J. S., Barrera Medina, R. A., Merino Molina, V., & Hernandez Rincon, E. H. (2026). Beyond the conventional: Artificial intelligence in identifying risk factors in sports injuries. A scoping review. International Journal of Medical Informatics. https://doi.org/10.1016/j.ijmedinf.2026.106408