JMIR Publications

ABSTRACT

Background:

Foot orthosis prescription is a complex clinical decision-making process that involves selecting and combining multiple structural, functional, and material components based on heterogeneous biomechanical information. In routine outpatient practice, detailed biomechanical assessments are often incomplete, creating substantial variability in prescription decisions and limiting the applicability of conventional machine learning (ML) models that assume fixed feature availability.

Objective:

This study aimed to develop and evaluate a hierarchical ML–based clinical decision support framework for foot orthosis prescription that accommodates variable clinical information availability, supports multi-label prescription decisions, and emphasizes clinical safety in real-world outpatient settings.

Methods:

A retrospective observational study was conducted using 6,462 visit-level clinical encounters collected during routine outpatient care between 2015 and 2020. Orthotic prescription was formulated as a multi-label prediction task involving 15 prescription components. A hierarchical modeling framework was implemented, consisting of a basic decision level using routinely available demographic and alignment variables and an advanced level incorporating subtalar joint inversion and eversion range of motion measurements when available. Tree-based gradient boosting models were evaluated using 5-fold stratified cross-validation. Model performance was assessed using component-wise area under the receiver operating characteristic curve (AUROC), macro-averaged Hamming loss, Top-K inclusion rates, and safety-oriented threshold calibration.

Results:

The hierarchical models demonstrated stable and clinically meaningful performance under heterogeneous data conditions. The Level 4 model achieved a mean component-wise AUROC of 0.815, while the Level 8 model achieved a higher mean AUROC of 0.824 with a lower macro-averaged Hamming loss (0.119 vs 0.110). Top-K analysis showed that clinician-prescribed components were captured in 86.5% of cases at Top-1 and 96.7% at Top-3, exceeding 99% at Top-8. Safety-oriented threshold calibration prioritized high specificity (95.12%), resulting in a conservative recommendation profile that minimized over-prescription while maintaining moderate recall (66.64%). Age-stratified analyses revealed distinct prescription determinant patterns between pediatric and adult populations, supporting the clinical plausibility of the learned decision rules.

Conclusions:

The proposed hierarchical clinical decision support framework provides robust, safety-oriented, and workflow-compatible support for multi-component foot orthosis prescription under real-world outpatient constraints. By adapting to variable information availability and generating interpretable, ranked recommendations, the system augments―rather than replaces―clinician judgment and bridges the gap between ML models and routine orthotic practice.