{"resourceId":"lindskog-sports-medicine-ai-maturity-review-2026","versions":[{"version":"external-908bcf40c786065b7981d71a229bfc989184eb115c0b969cc93d8df4d18ec43b","resource":{"id":"lindskog-sports-medicine-ai-maturity-review-2026","title":"Sports-medicine review finds external validation remains uncommon","organization":"Jakob Lindskog and colleagues; University of Gothenburg and collaborators","sector":"Sports medicine; transfer to collegiate athlete support","geography":"International literature; Sweden-led review","publishedAt":"June 12, 2026","publicationDate":"2026-06-12","eventDate":null,"sourceName":"BMC Medical Informatics and Decision Making","sourceLabel":"Peer-reviewed scoping review; methods and limitations inspected","sourceUrl":"https://link.springer.com/article/10.1186/s12911-026-03615-w","evidenceClass":"academic-research","outcomeClass":"cautionary","topics":["knowledge-work","data-security","governance-procurement","accessibility-workforce","operating-model"],"finding":"A review of 97 studies found only four using external datasets to assess generalizability; strong internal performance did not establish deployment readiness.","sledRelevance":"Independent scrutiny for collegiate sports-medicine procurement; populations extend beyond U.S. student-athletes.","evidence":"MEDLINE, EMBASE and Web of Science were searched February 5, 2026; two reviewers screened studies. The synthesis was narrative, without pooled effects or formal risk-of-bias appraisal. Prospective workflow testing was uncommon.","architectureImplications":"Interpretation: create a read-only evaluation path before connecting model output to clinical systems; log input timing and model versions.","governanceImplications":"Interpretation: medical leadership should set acceptable use and retain decision authority.","securityPrivacyImplications":"Interpretation: minimize linked health and performance records; segregate medical access from selection decisions.","caveats":"English-language scope, heterogeneous designs, no independent replication of included models and a February search cutoff. Not a NCAA-only assessment or proof every model fails.","streamIds":["college-athletics"],"roles":{"sales":"Interpretation: The customer problem is deciding whether an appealing prediction tool is suitable for an actual sports-medicine decision. Engage the team physician, athletic trainers, performance staff, IT and procurement. Ask what action a risk score would change, what the current baseline achieves, and whether a vendor has tested comparable athletes prospectively. Offer a bounded evidence and data-readiness assessment before any clinical pilot. The value hypothesis is avoiding unsuitable integration and focusing evaluation effort. The review supplies questions for diligence, not a local injury-reduction estimate. Do not equate a high benchmark score with safer return-to-play decisions or guaranteed reductions in care costs.","engineering":"Interpretation: Fit is a clinician-supervised validation environment using authorized representative data. Prerequisites include clear outcome definitions, timestamped features, an approved comparison process and access to model documentation. Prevent leakage by separating athletes and future periods appropriately; assess calibration, subgroup errors and decision consequences rather than accuracy alone. Test missing inputs, stale readings and distribution changes. Treat any LLM explanations as a separate component needing factual review. Cloud, local and hybrid designs all require enforceable health-data boundaries. A proof of value should remain disconnected from treatment changes until medical and institutional review approve a prospective protocol.","delivery":"Interpretation: The sports-medicine director should own the decision workflow, with athletic trainers, a statistician and data engineers supporting validation. Establish the baseline, document data quality, run a shadow assessment and review errors before proposing supervised use. Dependencies include consent or other institutionally approved authority, reliable records and protected staff time. Train users to recognize uncertain outputs and record overrides. Governance checkpoints should approve data intake, evaluation design and model updates. Proposed acceptance criteria include complete provenance, no critical access failures, documented subgroup evaluation and medical sign-off on any progression. Small samples and changing practice patterns can invalidate apparently favorable results."},"retrievedAt":"2026-09-09T03:01:52Z","enrichedAt":"2026-09-09T03:04:23Z","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Interpretation: test explanations with athletes and clinicians, including accessible alternatives to dashboards.","procurementImplications":"Interpretation: require evidence relevant to the intended population, decision and workflow, plus exportable evaluation records.","operatingModelImplications":"Interpretation: appoint a clinical owner, an evaluation lead and explicit model-change approval.","updateExplanation":"New archive source supplying independent validation scrutiny for collegiate performance-technology decisions. June publication and February search cutoff are preserved; no new-since-yesterday claim.","sourceVerification":{"openedUrl":"https://link.springer.com/article/10.1186/s12911-026-03615-w","referenceExcerpt":"Only four studies employed external datasets to assess generalizability","promptVersion":"sled-research-v3.1","model":null,"basis":"agent-reported inspection"}}}]}