{"resourceId":"claros-post-concussion-msk-ml-2025","versions":[{"version":"external-955bb37899dd3f108268f07a1f3d7c476cae8193979f016cf4a811fe642e4d5a","resource":{"id":"claros-post-concussion-msk-ml-2025","title":"Claros et al.: Collegiate post-concussion MSK injury ML model reports AUC 0.82","organization":"University of Delaware / Sports Medicine (Springer)","sector":"College athletics / sports medicine and performance risk modeling","geography":"United States (single NCAA institution; CARE-aligned assessments)","publishedAt":"March 27, 2025","publicationDate":"2025-03-27","eventDate":null,"sourceName":"Sports Medicine (PMC full text)","sourceLabel":"Academic research — measured post-concussion injury risk model","sourceUrl":"https://pmc.ncbi.nlm.nih.gov/articles/PMC12460441/","evidenceClass":"academic-research","outcomeClass":"mixed","topics":["data-security","knowledge-work","operating-model","infrastructure"],"finding":"Using 194 NCAA student-athletes (155/39 train/test) and 135 candidate variables from demographics, concussion characteristics, and standard clinical assessments, Claros et al. report a composite ML risk score with AUC 0.82; at a 6.67% false-positive rate, sensitivity was 79% and precision 95%. Top predictors included cognitive, balance, and reaction measures. Results support identifying elevated post-concussion MSK risk for targeted prevention—not a multi-site product deployment.","sledRelevance":"Direct collegiate sports-medicine evidence for athletic trainers and performance staff considering AI risk stratification after concussion, with explicit metrics and CARE/SCAT-aligned inputs transferable to other NCAA programs that already collect these assessments.","evidence":"PMC full text of Sports Medicine article (doi:10.1007/s40279-025-02196-4): methods (WoE transform, L1-penalized logistic selection, AIC), AUC/sensitivity/precision figures, Key Points, and single-institution CARE Consortium enrollment context.","architectureImplications":"Interpretation: integrate risk scoring with existing concussion EMR/assessment pipelines rather than a parallel wearable stack; require versioned feature dictionaries for SCAT/CARE fields and held-out calibration monitoring before coach-facing alerts.","governanceImplications":"Interpretation: treat scores as clinical decision support under sports-medicine ownership; define who may see risk tiers (AT vs coach) and prohibit scholarship/roster use without separate policy.","securityPrivacyImplications":"Interpretation: concussion and cognitive assessment data are highly sensitive; limit secondary AI training rights and apply minimum-necessary access even inside athletics.","caveats":"Single institution; test set n=39; no published multi-site external validation; not an evaluation of a commercial product; secondary press may overstate '95% accuracy.'","streamIds":["college-athletics"],"roles":{"sales":"Open with post-concussion return-to-play risk, not generic 'injury AI.' Athletic trainers already capture SCAT/CARE-like measures; the credible ask is whether a calibrated risk score can prioritize prevention workloads without promising injury elimination. Discover who owns concussion data, whether coaches see cognitive scores today, and what false-positive burden ATs will tolerate. Bound value to better targeting of risk-reduction programs on standard assessments—not to win totals or roster automation.","engineering":"Fit is clinical decision-support on existing assessment exports, not a new sensor fleet. Prefer batch or EHR-adjacent scoring with documented feature schemas, calibration plots, and site-specific revalidation before any coach dashboard. Prerequisites include clean longitudinal concussion records, IRB/clinical governance, and a holdout policy. Constraints: single-site training distribution and small test n. Proof of value is local AUC/PPV at an agreed FPR plus clinician override logs—not vendor accuracy slides.","delivery":"Delivery is sports-medicine led: map assessment fields, validate on local historical cases, set alert thresholds with ATs, train staff that scores inform not replace clearance, and monitor drift each season. Dependencies include AT documentation quality and IT access to assessment stores. Ownership sits with sports medicine under athletic-director sponsorship. Acceptance criteria: documented local validation, written coach-access rules, and zero unauthorized scholarship use of scores in the pilot window."},"roleDetails":{"sales":{"customerProblem":"After concussion, collegiate athletes face elevated MSK injury risk, but staff lack a validated way to prioritize who needs intensified prevention.","stakeholders":["Head athletic trainer","Team physician","Sports performance director","Sport coaches","Compliance/privacy officer"],"discoveryQuestions":["Which concussion assessments are already mandated and stored?","Who may view cognitive or balance scores today?","What false-positive rate is acceptable for AT workload?","Would risk tiers ever influence scholarship or roster decisions?"],"valueHypothesis":"If local revalidation holds near the paper's discrimination, ATs could target prevention programs to higher-risk athletes using data they already collect—without claiming injury elimination.","potentialEngagement":"Six-to-eight-week retrospective validation on the school's last two seasons of concussion cases, plus a draft clinical-use policy.","unsupportedClaims":["Guaranteed reduction in post-concussion MSK injuries","Multi-site generalizability without local validation","That 95% is overall accuracy rather than precision at a chosen FPR"]},"engineering":{"fit":"Clinical decision-support scoring on existing concussion assessment pipelines for NCAA sports medicine.","architectureIntegration":"Export SCAT/CARE-aligned fields to a versioned feature store; score in a controlled environment; return risk tiers to AT workflows with audit logs.","prerequisites":["Longitudinal concussion and subsequent MSK injury labels","Standardized assessment timing","Clinical governance approval"],"constraints":["Single-institution training distribution in the source study","Small held-out test set","Missing-data and categorical encoding sensitivity"],"security":"Restrict cognitive/concussion features; no secondary model-training rights for vendors without explicit athlete-facing consent review.","proofOfValue":["Local AUC and PPV at agreed FPR on held-out seasons","Calibration within agreed error bands","Clinician override rate reviewed monthly"]},"delivery":{"work":"Historical validation, threshold setting with ATs, coach-access policy, and season-one monitoring dashboard.","dependencies":["Athletic training documentation quality","IT access to assessment and injury systems","Team physician sign-off"],"ownership":"Sports medicine / head athletic trainer owns clinical use; AD sponsors policy.","skillsAdoption":"ATs learn score interpretation and limits; coaches learn they do not receive raw cognitive detail without policy.","governanceCheckpoints":["IRB/clinical review before prospective use","Written prohibition on scholarship use","Quarterly drift review"],"acceptanceCriteria":["Documented local validation report","Signed access matrix","No unauthorized roster uses in pilot"],"risks":["Over-trust in precision headline figures","Leak of cognitive scores to coaching staff","Distribution shift across sports"]}},"retrievedAt":"2026-09-20T03:13:05Z","enrichedAt":"2026-09-20T03:15:40Z","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Interpretation: ATs need plain-language score cards and workload estimates at chosen alert thresholds so junior staff are not flooded by false positives.","procurementImplications":"Interpretation: buy validation and integration services around existing assessment vendors; do not procure a black-box 'injury AI' that ignores SCAT/CARE field definitions.","operatingModelImplications":"Interpretation: keep clearance and prevention planning with sports medicine; models triage attention, they do not clear athletes.","sourceVerification":{"openedUrl":"https://pmc.ncbi.nlm.nih.gov/articles/PMC12460441/","referenceExcerpt":"This study developed a composite risk score model with an area under the curve of 0.82","promptVersion":"sled-research-v4.0","model":null,"basis":"agent-reported inspection"}}}]}