{"resourceId":"palermi-diagnostics-ai-sports-medicine-framework-2026-05-09","versions":[{"version":"external-cde4f6ee0f17c330a659d74f1580aa3708474c905d01c8fa1f7cc8ffdd085f3b","resource":{"id":"palermi-diagnostics-ai-sports-medicine-framework-2026-05-09","title":"Diagnostics narrative review: sports physicians should keep AI as a decision-support layer with human-in-the-loop return-to-play authority","organization":"UniCamillus and multi-institution coauthors / Diagnostics (MDPI)","sector":"Sports medicine clinical practice","geography":"International (Italy/US/France-centered author team; literature global)","publishedAt":"May 9, 2026","publicationDate":"2026-05-09","eventDate":null,"sourceName":"Diagnostics (MDPI)","sourceLabel":"Peer-reviewed narrative review","sourceUrl":"https://www.mdpi.com/2075-4418/16/10/1448","evidenceClass":"academic-research","outcomeClass":"cautionary","topics":["governance-procurement","operating-model","data-security","accessibility-workforce"],"finding":"Palermi et al. present a structured narrative review (PubMed/MEDLINE 2015–2024; about 1284 records screened to 74 included) arguing AI is reshaping sports-medicine decisions across injury risk, MSK imaging, rehab, return-to-play, and workflow, but should remain a probabilistic decision-support layer. They report injury-model AUCs commonly about 0.70–0.85 with limited external validity, warn that FDA 510(k) clearance is not athlete-specific outcome proof, and center the sports physician as human-in-the-loop integrator for high-stakes RTP. Conflicts disclosed include industry ties for some coauthors (e.g., Gleamer).","sledRelevance":"Not in archive; newly relevant with same-edition Huang SLR and Duan elite causal paper as clinical-governance counterweight for collegiate sports-medicine AI purchasing after recent injury-model archive themes.","evidence":"Full article opened via HTTPS WebFetch (direct curl received 403). Abstract/conclusions state decision-support framing and human-in-the-loop role; methods describe narrative synthesis of 74 studies; sections detail imaging devices table and competency framework. Collegiate-specific trials are not the focus; transfer requires local governance.","architectureImplications":"Favor architectures that surface uncertainty, preserve clinician override, and integrate multimodal inputs into EHR/team workflows without autonomous RTP clearance.","governanceImplications":"Codify that AI outputs cannot alone clear RTP or determine scholarships/playing time; require bias monitoring for underrepresented athlete groups.","securityPrivacyImplications":"Continuous wearable/imaging/EHR fusion heightens ownership, secondary use, and commercialization risks the authors flag—contracts must follow.","caveats":"Narrative not systematic meta-analysis; search ended Dec 2024; some authors have AI-imaging industry roles; not NCAA-regulation guidance.","streamIds":["college-athletics"],"roles":{"sales":"Sales framing: Team physicians need AI that speeds imaging and load synthesis without ceding return-to-play authority. Use this review’s human-in-the-loop standard to sell clinician-centered decision support with override UX, uncertainty display, and population-matched validation—not autonomous clearance. Bound engagements to one workflow under sports-medicine governance. Unsupported: that FDA 510(k) clearance proves athlete outcome benefit, that AI can replace physician RTP judgment, or that performance is uniform across women, youth, and para athletes underrepresented in many training sets.","engineering":"Engineering framing: Fit is clinical decision support and imaging/rehab analytics with mandatory human confirmation for high-stakes states. Architecture should stream read-only model outputs into EHR/team workflows, log overrides, and never auto-flip clearance. Prerequisites: intended-use alignment, PACS/EHR connectors, and role-based access. Constraints include limited athlete-specific validation, automation bias, and LLM summarizer hallucination risk. Security protects multimodal health data and bans undisclosed secondary commercial use. Proof of value: time-to-relevant-context and stable clinician agreement—not model AUC alone.","delivery":"Delivery framing: Pilot AI imaging or load summaries under a sports-medicine SOP that requires physician confirmation on every high-stakes decision. Dependencies include privacy/counsel review, vendor intended-use checks, and athlete communication about data use. Skills adoption builds AI appraisal, uncertainty communication, and bias checks into medical staff training. Governance checkpoints: pre-go-live ethics review, 30-day override audit, and equity performance checks across sports. Acceptance: 100% high-stakes decisions documented as physician-final. Risks: automation bias in busy clinics and tool access limited to revenue sports."},"roleDetails":{"sales":{"customerProblem":"Sports medicine groups face AI vendor pressure without a clear rule that physicians retain high-stakes decisions.","stakeholders":["Head team physician","Athletic trainer leadership","AD","University counsel/privacy"],"discoveryQuestions":["Which decisions will AI inform versus decide?","How are overrides logged?","What athlete groups are in the training data?"],"valueHypothesis":"Clinician-centered AI support may improve information synthesis if RTP and selection authority stay human—and if validation matches the use case.","potentialEngagement":"Policy + pilot design for one AI imaging or load-summary tool under sports-medicine governance.","unsupportedClaims":["AI replaces team physician RTP judgment","510(k) clearance proves athlete outcome benefit","Uniform performance across all athlete populations"]},"engineering":{"fit":"Decision-support CDSS/imaging/rehab analytics with mandatory human confirmation.","architectureIntegration":"Read-only model outputs into clinical workflow; override capture; no automated clearance state changes.","prerequisites":["Intended-use alignment","EHR/PACS connectors","Uncertainty UX","Access control"],"constraints":["Limited athlete-specific validation","Automation bias","LLM hallucination risk"],"security":"Protect multimodal health data; ban secondary commercial use; breach notification to athletes where required.","proofOfValue":["Override/edit rates","Time-to-relevant-context","Disagreement cases reviewed weekly"]},"delivery":{"work":"Write AI-in-clinic SOP, train physicians/ATCs, pilot one workflow, and audit overrides before expansion.","dependencies":["Vendor clinical validation dossier","Counsel/privacy","Change management time"],"ownership":"Head team physician owns clinical decisions; athletics IT owns integrations.","skillsAdoption":"Build AI literacy, uncertainty communication, and bias checks into medical staff training.","governanceCheckpoints":["Pre-go-live ethics/privacy review","30-day override audit","Equity performance check"],"acceptanceCriteria":["100% high-stakes decisions documented as physician-final","No autonomous RTP flag in production"],"risks":["Automation bias","Vendor conflict influence","Widening care gaps if only elite sports funded"]}},"retrievedAt":"2026-10-02T03:17:30Z","enrichedAt":"2026-10-02T03:29:00Z","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Framework demands new clinician AI-appraisal skills; under-resourced sports may lack training time, widening tool disparities.","procurementImplications":"Purchase only tools with clear intended use, override design, and athlete-population validation—not marketing accuracy alone.","operatingModelImplications":"Reinforces physician-final operating model for AI-augmented sports medicine in collegiate programs.","sourceVerification":{"responseId":"webfetch-mdpi-palermi-20261002T031730","openedUrl":"https://www.mdpi.com/2075-4418/16/10/1448","referenceExcerpt":"AI is likely to become an important decision-support layer in sports medicine","promptVersion":"sled-research-v4.0","model":null,"basis":"agent-reported inspection"}}}]}