AHN-BudgetNet: Cost-Aware Multimodal Feature-Acquisition Architecture for Parkinson’s Disease Monitoring

This work presents AHN-BudgetNet, a novel cost-aware, tiered feature-acquisition framework employing hierarchical organization and systematic cost-effectiveness optimization for quantifying marginal predictive value and economic efficiency in multimodal clinical assessments for Parkinson’s disease motor severity prediction. Trained and validated on 1,387 baseline subjects from the PPMI cohort using patient-level GroupKFold cross-validation, our framework evaluates 31 possible tier combinations across five hierarchical assessment categories: demographics (Tier 0, $0), patient-reported outcomes (Tier 1, $75), clinical evaluations (Tier 2, $300), neuroimaging biomarkers (Tier 3, $3,300), and advanced molecular biomarkers (Tier 4, $5,000). The hierarchical framework systematically evaluates tier-level importance through Random Forest classification with comprehensive cross-validation, enabling efficient feature selection across cost strata. Results demonstrate that demographics alone (single feature: age) achieve baseline performance (AUC: 0.503, efficiency: 5.03), while patient-reported assessments provide exceptional cost-effectiveness (AUC: 0.802, efficiency: 4.58) at minimal investment ($75). The optimal performance combination (T0+T1+T2) achieves AUC 0.847 at $375 total cost with efficiency 1.78, representing 69% improvement over baseline demographics with 5-fold cost increase. High-cost modalities (Tiers 3-4) showed complete unavailability (100% missing) or severe data sparsity (88.6-90.5% missing), validating real-world implementation constraints and supporting the tiered acquisition strategy. Spectral clustering on Tier 1 features achieved optimal patient stratification (silhouette score: 0.654), enabling personalized monitoring protocols. The framework provides evidence-based decision rules through cost-effectiveness optimization: $75 budget scenarios should utilize Tier 1 assessments, while $375 budgets justify comprehensive T0+T1+T2 evaluation. This hierarchically structured framework offers a practical approach for resource-constrained clinical decision-making in neurodegenerative disease monitoring, demonstrating that systematic tiered assessment prioritization maintains diagnostic accuracy while optimizing healthcare resource utilization and supporting precision medicine implementation in diverse economic contexts.