{"resourceId":"etamu-ncaa-di-dxa-dari-injury-risk-ml-2026","versions":[{"version":"external-6426832385d0363c7663ad1b9ae7e4be1bc989a952f25a0d7c17507321952be9","resource":{"id":"etamu-ncaa-di-dxa-dari-injury-risk-ml-2026","title":"East Texas A&M abstract: DXA and DARI screening ML predicts 180-day injury risk in 1,258 NCAA DI athlete records with AUC 0.74","organization":"East Texas A&M University ROARHP Lab (Strogalev, Bernhardt, Oldham et al.)","sector":"NCAA Division I sports science / athlete health","geography":"United States (Commerce, Texas)","publishedAt":"2026 (Texas ACSM abstract, Int J Exerc Sci Conference Proceedings Vol 2 Iss 18; exact date not stated)","publicationDate":null,"eventDate":null,"sourceName":"International Journal of Exercise Science: Conference Proceedings (TACSM abstracts)","sourceLabel":"Academic conference abstract","sourceUrl":"https://digitalcommons.wku.edu/cgi/viewcontent.cgi?article=9672&context=ijesab","evidenceClass":"academic-research","outcomeClass":"mixed","topics":["data-security","governance-procurement","operating-model"],"finding":"A retrospective two-stage CatBoost model using DXA body composition, DARI markerless motion analysis, and Y-Balance data from 1,258 NCAA DI athlete records (246 injuries within 180 days; scans Aug 2022 to Dec 2025) reached AUC 0.74 on a hold-out set for any injury. Body-region prediction among injured athletes had 50.0% top-1, 62.5% top-2, and 77.1% top-3 accuracy. The authors note single-site prediction remains challenging given class imbalance and limited samples.","sledRelevance":"Why now: Ohio State announced Project SCARLET (Sep 29 to Oct 1, 2026), using predictive models across 36 sports; this abstract gives a concrete, modest benchmark for what screening-based injury prediction achieves in Division I data.","evidence":"Master’s-category conference abstract; retrospective single-institution dataset; hold-out testing but no external validation, calibration, or prospective test reported. AUC 0.74 for binary injury risk; body-region top-k accuracies as stated. No sport-by-sport breakdown or intervention outcome.","architectureImplications":"Combines DXA, DARI motion capture, and balance testing into a model pipeline; requires linked injury records from sports medicine. Interpretation: data integration across performance and medical systems is the main build effort.","governanceImplications":"Injury-risk scores touch medical decisions; departments should keep them as decision support under sports-medicine authority with documented thresholds and athlete communication.","securityPrivacyImplications":"DXA scans, motion data, and injury records are sensitive health information; linking them requires consent, access controls, and retention rules.","caveats":"Conference abstract, not peer-reviewed full paper; single institution; retrospective; publication date not stated beyond 2026; AUC 0.74 is moderate discrimination.","streamIds":["college-athletics"],"roles":{"sales":"Sales framing: the conversation is about realistic expectations for injury-risk AI. This East Texas A&M abstract shows moderate discrimination (AUC 0.74) from screening data already collected by many programs. A bounded engagement helps a department test such a model on its own data before purchase or rollout. Avoid claims that models prevent injuries or pinpoint injury sites; the abstract shows neither.","engineering":"Engineering framing: fit depends on linking DXA, DARI, balance, and injury records under one athlete identifier. Engineers should check data completeness, define the injury label and window, use time-based hold-outs, and report calibration and subgroup performance by sport and sex. External or prospective validation is the necessary next step before operational use. Engineers should also confirm that screening protocols stayed consistent across the years used for training.","delivery":"Delivery framing: run injury-risk modeling as a sports-medicine-owned pilot. Delivery should set data-sharing agreements between performance and medical staff, define how scores reach clinicians, track calibration over a season, and review whether flagged athletes received interventions. Athlete consent and communication plans are prerequisites. The pilot should run for at least one full season so calibration and intervention follow-through can be judged on real outcomes rather than retrospective fit, and results should be reviewed jointly by sports medicine and performance staff."},"roleDetails":{"sales":{"customerProblem":"Programs collect screening data but lack evidence on how well it predicts injuries.","stakeholders":["Head team physician","Director of sports medicine","Strength and conditioning director","Sports science staff","Compliance/privacy officer"],"discoveryQuestions":["Do you link DXA and motion screening to injury records?","What predictive performance would justify clinical use?","Who decides on interventions from a risk score?","What athlete consent covers this use?"],"valueHypothesis":"Local validation of screening-based models can show whether risk scores add useful information to clinicians.","potentialEngagement":"A retrospective validation study on the department’s own screening and injury data.","unsupportedClaims":["Injury prevention or reduced injury rates","Accurate single-site injury prediction","Generalization beyond one institution","Clinical-grade decision support"]},"engineering":{"fit":"Programs already running DXA and DARI or similar screening with structured injury records.","architectureIntegration":"ETL from DXA, DARI, and balance tests plus injury database into a model pipeline with hold-out evaluation.","prerequisites":["Consistent athlete IDs across systems","Defined injury taxonomy and 180-day window","Data-sharing agreement with sports medicine"],"constraints":["Class imbalance","Single-site data","Changes in screening protocol over time"],"security":"Treat all inputs as health data; role-based access; de-identify for model development.","proofOfValue":["Time-based hold-out AUC and calibration","Subgroup performance by sport and sex","Prospective one-season test"]},"delivery":{"work":"Pilot an injury-risk model under sports-medicine governance with prospective tracking.","dependencies":["Medical and performance data integration","Consent updates","Clinical review process"],"ownership":"Sports medicine leadership, with sports science support.","skillsAdoption":"Train clinicians and coaches on interpreting moderate-accuracy risk scores.","governanceCheckpoints":["IRB or privacy review","Threshold and action-plan approval","End-of-season calibration review"],"acceptanceCriteria":["Proposed: prospective AUC and calibration reported after one season","Proposed: documented clinician review of every flagged athlete"],"risks":["Over-trusting moderate predictions","Misuse in playing-time or roster decisions","Privacy breaches of health data"]}},"retrievedAt":"2026-10-04T22:23:45-05:00","enrichedAt":"2026-10-04T22:27:00-05:00","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Requires sports-science analytic capacity; clinicians need training on probabilistic outputs.","procurementImplications":"Vendors offering injury prediction should provide external validation and calibration; buyers should require local validation rights.","operatingModelImplications":"Bridges performance and sports medicine data; governance must keep medical authority over any intervention.","sourceVerification":{"responseId":"webfetch-wku-etamu-injury-20261004T222345","openedUrl":"https://digitalcommons.wku.edu/cgi/viewcontent.cgi?article=9672&context=ijesab","referenceExcerpt":"The injury risk prediction model achieved an area under the receiver operating characteristic curve (AUC) of 0.74","promptVersion":"sled-research-v4.0","model":null,"basis":"agent-reported inspection"}}}]}