From the College Athletics edition of September 30, 2026
Hando et al. MSSE response: current sports-medicine AI/ML lacks real-world validation for clinical use
Medicine & Science in Sports & Exercise (ACSM) · Sports medicine / exercise science (athlete and military populations) · United States (UVA / Baylor–Army authors; military and athletic deployment contexts)
- Publisher
- Medicine & Science in Sports & Exercise
- Original publication
- Publish Ahead of Print (MSSE 2026); discussed in Newswise 2026-06-23
- Source retrieved
- 2026-10-01
What happened
In an MSSE ahead-of-print response (DOI 10.1249/MSS.0000000000004046), Hando, Kupperman, Brown, and Angadi argue current AI/ML systems used by athletes, military units, and health organizations have not undergone requisite real-world testing; where externally validated, outputs were inaccurate or misleading. They stress transparency/explainability/accountability, criticize black-box wearables and circular validation, note high LLM hallucination rates, and conclude evidence does not support clinical AI/ML use in sports medicine pending rigorous research. Table contrasts optimistic claims with limited study utility (e.g., injury model true-positive rate 53.7% without prospective validation).
Why it matters
Why now: Directly counters the deployment narrative around collegiate injury/readiness ML highlighted by the 2026-09-28 BMC high-accuracy Kaggle study and ongoing vendor injury-AI marketing. Provides peer-reviewed sports-medicine scrutiny athletics clinicians can cite in procurement gates.
Evidence and measured results
Full PDF inspected: authors state disagreement that AI has arrived for clinical use; cite their prior military cohort validations showing poor utility; discuss black-box circular validation; GPT-5 hallucination range 26–47% cited; Table 1 critiques imaging/injury/digital-twin claims; recommend external validation, adversarial testing, monitoring. Military-focused empirical base; collegiate use referenced as deployment context in related coverage.
Limitations and uncertainty
Response essay, not a new multi-school RCT; strongest empirical citations are military cohorts; exact print date not on PDF (ahead of print); collegiate-specific metrics limited.
Put this evidence to work
Lighthouse Advisory interpretation, grounded in this source. Enriched 2026-10-01; this does not change the original publication date. Each role follows the same checklist so takeaways compare across sources.
Sales
Role takeaway
- Customer problem
- Vendors sell injury-risk and readiness AI to colleges without the external validation sports-medicine reviewers say is required for clinical use.
- Stakeholders
- Team physicians
- ATCs
- Sports performance
- AD / risk management
- Discovery questions
- Has the tool been prospectively validated in a similar population?
- What happens when the score disagrees with the clinician?
- Can we independently audit the model?
- Value hypothesis
- Validation-first sports-medicine AI advisory reduces clinical and reputational risk versus buying opaque injury predictors on marketing claims.
- Potential engagement
- Sports-medicine AI inventory and evidence review against SaMD-like criteria for one high-risk tool.
- Unsupported claims
- Guaranteed injury prevention
- That military validation results transfer unchanged to every NCAA sport
- That LLMs safely generate physiological hypotheses
Pre-sales engineering
Role takeaway
- Fit
- Clinical decision-support governance for athlete health AI; reject black-box injury oracles without prospective proof.
- Architecture and integration
- Shadow scoring with clinician override; model cards; monitoring for drift; no closed-loop RTP automation.
- Prerequisites
- Prospective outcomes pipeline
- Independent validation cohort plan
- Explainability requirements in vendor RFP
- Constraints
- Noisy wearable ground truth
- Rare injuries
- Proprietary vendor refusal to open models
- Security
- Audit logs for score-influenced decisions; prohibit undisclosed secondary model training on athlete data.
- Proof of value
- Prospective sensitivity/specificity
- Calibration plots
- Adversarial/external test results
Delivery
Role takeaway
- Work
- Pause clinical reliance on unvalidated tools; design prospective evaluation or decommission; train clinicians on failure modes.
- Dependencies
- Sports medicine authority
- Vendor data access for audit
- IRB/privacy if researchifying evaluation
- Ownership
- Head team physician / sports medicine medical director.
- Skills and adoption
- Clinician literacy on external validation vs marketing AUC.
- Governance checkpoints
- Purchase gate with evidence dossier
- Quarterly performance monitoring
- Incident review when scores mislead
- Acceptance criteria
- No unvalidated score as sole RTP determinant
- Documented external validation or tool removal
- Risks
- False reassurance
- Unnecessary restriction of athletes
- Liability from opaque recommendations
Implementation considerations
Lighthouse Advisory interpretation across the operating dimensions a public-sector buyer must settle before this evidence becomes a design. Each note answers the question under its heading for this specific source.
Architecture and integration
What must connect, and where does the AI sit in the workflow?
Do not wire unvalidated injury-risk or readiness models into return-to-play decision systems; prefer transparent models with prospective monitoring and human clinical authority.
Governance
Who approves, reviews and stays accountable for outcomes?
Apply SaMD-like expectations (transparency, real-world performance monitoring, pre-use validation) to sports-science software affecting athlete health decisions.
Security and privacy
What data, permissions and controls need testing?
Black-box vendor models processing wearable/biomechanical data impede auditability; contracts should require explainability and independent evaluation rights.
Accessibility and workforce
Who is affected, and what skills or accommodations follow?
Keeps clinicians and experienced practitioners central; warns humans-with-unvalidated-AI can be worse than trained humans without it.
Procurement
What should contracts, pricing and exit terms secure?
RFPs for injury/readiness software should demand peer-reviewed external validation, adversarial testing, and real-world monitoring—not utilization rates.
Operating model
Which teams own the service once it runs?
Sports medicine should treat most commercial AI as experimental until prospectively proven, with clear clinical ownership.
Publication history
- 2026-09-30College Athletics · Issue 235 resources
Stable resource ID: hando-msse-ai-not-ready-sports-medicine-2026