A safety-gated, budget-constrained algorithm can recommend footwear with clinically acceptable expert concordance

Claim

Among adults without urgent foot-related red flags, a transparent rule-based algorithm that considers foot width, pain location, standing time, walking surface, orthotic use, donning difficulty and budget can recommend footwear attributes and specific models with at least 80% agreement with an independent multidisciplinary expert panel. Rationale Footwear selection depends on anatomy, function, surface, symptoms, safety risks and affordability. Existing evidence supports specific footwear characteristics in defined populations, but it does not validate a universal product recommendation system. The algorithm therefore separates: A favorable compatibility score cannot override a safety hold. Proposed methodology Evaluate the algorithm prospectively using at least 150 standardized cases spanning different foot widths, pain regions, occupational demands, surfaces, orthotic requirements, diabetes-related foot risks and budgets. Each case will be reviewed independently by a rheumatologist, rehabilitation specialist and foot-care professional blinded to the algorithm output. Primary outcomes: Secondary outcomes:

1. Clinical safety screening.
2. Functional compatibility.
3. Economic eligibility.
4. Product-level evidence.
5. Unknown or unverifiable specifications.
   1. Agreement between the algorithm and expert consensus on the preferred footwear profile.
   2. Proportion of unsafe recommendations in cases requiring professional assessment.
   3. Compliance with the stated budget.
   4. Agreement on the highest-ranked specific model when sufficient verified product data exist.
   • Inter-rater agreement among experts.
   • Recommendation stability after catalog updates.
   • Explanation completeness. (1/2)

Falsification criteria The claim will be rejected if: Current evidence status The deployed prototype demonstrates deterministic execution, safety holds, budget filtering, source-linked product records and explicit handling of unknown data. It has not yet established clinical concordance or improved patient outcomes. Relevant evidence: Prototype: https://rheumascore.xyz/exp-footwear-ai.html

1. Frequency of abstention caused by missing product data.
2. Accessibility across economic groups.
   • Expert concordance is below 80%.
   • The algorithm generates any product recommendation after a prespecified safety hold.
   • More than 5% of recommendations exceed the selected budget.
   • Product rankings depend on specifications that cannot be traced to an exact model and source.
   • IWGDF guidance on diabetes-related foot risk: PMID 37302121
   • JOSPT guideline for plantar heel pain: PMID 38037331
   • Randomized footwear trial in knee osteoarthritis: PMID 33428439