Budgeted Human Review for Privacy-Aware Clinical and Financial Risk Triage

Xuanyang Lu

Budgeted Human Review for Privacy-Aware Clinical and Financial Risk Triage

Authors

Xuanyang Lu Computer Science, Johns Hopkins University, MD, USA

Keywords:

human review, risk triage, differential privacy, calibration; fairness, clinical decision support, credit assessment

Abstract

Risk systems must decide not only which cases are high risk, but also which cases deserve scarce human review. This paper studies a budgeted triage policy that combines calibrated risk scores, uncertainty, subgroup safeguards, and a fixed review capacity. A deterministic synthetic simulation represents clinical and financial decision streams with different base rates, asymmetric error costs, and institution-level distribution shifts. It compares fixed-threshold automation, uncertainty-only referral, cost-sensitive referral, and a privacy-budgeted fairness-aware policy. The simulated results show that review allocation based on expected avoidable harm captures more high-cost errors than a score-only queue under the same capacity, while explicit group caps prevent one subgroup from absorbing disproportionate deferrals. Stronger privacy noise reduces ranking resolution and can lower review yield, especially in small sites. The numerical results are generated by the stated simulator and are not estimates from patient or applicant data. The paper contributes a decision protocol that treats review capacity, privacy expenditure, calibration, and group-specific queue burden as coupled constraints.

References

[1]Yu, M., Zheng, Y., & Zhang, X. (2026). The Effect of Resampling Strategies on Tabular Fraud Detection Performance: A Comparative Empirical Evaluation. Annals of Applied Sciences, 7(1).

[2]Wang, J., Zheng, Y., & Zhang, X. (2026). A Comparative Empirical Study of Machine Learning Methods for Credit Card Fraud Detection: Gradient Boosting, Deep Tabular Learning, and Classical Baselines Under Extreme Class Imbalance. Spectrum of Research, 6(1).

[3]Wang, J., Zheng, Y., & Zhang, X. (2026). A Comparative Empirical Study of Machine Learning Methods for Credit Card Fraud Detection: Gradient Boosting, Deep Tabular Learning, and Classical Baselines Under Extreme Class Imbalance. Spectrum of Research, 6(1).

[4]Yu, M., Zheng, Y., & Zhang, X. (2026). The Effect of Resampling Strategies on Tabular Fraud Detection Performance: A Comparative Empirical Evaluation. Annals of Applied Sciences, 7(1).

[5]Li, Y., & Mo, Z. (2026). Does a Predicted 0.9 Mean 90 Percent? A Comparative Evaluation of Post-hoc Probability Calibration Methods for Default-Probability Scoring under Class Imbalance. Academia Nexus Journal, 5(2).

[6]Li, Y., & Mo, Z. (2026). Backtesting Value-at-Risk and Expected Shortfall Across Calm and Turbulent Markets: An Empirical Comparison of Historical Simulation, GARCH-Family, Filtered Historical Simulation, Extreme Value Theory, and CAViaR. Annals of Applied Sciences, 7(1).

[7]Chen, X. (2026, May). Feature Engineering and Algorithm Comparison for AI-Driven Payment Fraud Detection with Imbalanced Data. In Proceedings of the 2026 6th International Conference on Internet of Things and Machine Learning (pp. 662-671).

[8]Zhang, Y., & Hao, C. (2026). Comparative Empirical Evaluation of Monte Carlo Dropout Uncertainty Measures for Failure Capture and Referral Trade-offs in Chest X-ray Disease Detection: A Reproducible Simulation Study. Journal of Innovative Technologies, 2(2).

[9]Hu, L., & Hao, C. (2026). Confidence-Gated Arbitration of Automated Control Actions for Stability under Compounded Failures in High-Concurrency Platforms: A Comparative Empirical Study. Journal of Artificial Intelligence Review, 7(2), 109-118.

[10]Pan, H., & Zhong, M. (2026). Concept-Drift Detection for Systemic-Risk Early Warning: A Comparative Evaluation of Regime-Conditional Adaptive Thresholding under Distribution Shifts in Financial Stress Indices. Annals of Applied Sciences, 7(1).

[11]Wang, Y., & Mo, Z. (2026). Gradient-Boosted Trees Versus Classical Time-Series Baselines for 7-to 28-Day-Ahead Emergency-Department Respiratory Visit Pressure During Viral Co-Circulation: A County-Level Comparative Evaluation. Annals of Applied Sciences, 7(1).

[12]Wang, Y., & Tang, R. (2026). Comparing Unsupervised Spatial Grouping and Supervised Risk Ranking for County-Level Measles Susceptibility Stratification under Uneven Childhood Immunization Coverage. Journal of Selected Topics in Academic Research, 2(1).

[13]Liu, X. (2026, March). Interpretable Multi-horizon Risk Stratification for Alzheimer's Disease: A Clinical Feature-Based Approach Without Neuroimaging. In Proceedings of the 2026 6th International Conference on Bioinformatics and Intelligent Computing (pp. 535-540).

[14]Zhang, H., & Shi, W. (2026). Comparative Evaluation of Automated Detection Approaches for Identifying Implicit Compliance Violations in Cross-border Commercial Contract Clauses. Journal of Artificial Intelligence Review, 7(2), 1-22.

[15]Shang, W., Zhao, F., & Liu, M. (2026). A Comparative Study of Spatiotemporal Clustering and Classification Approaches for Security Incident Risk Assessment in UN Peacekeeping Operations. Journal of Sustainability, Policy, and Practice, 2(3), 1-11.

[16]Chen, X., Fu, H., & Pan, H. (2026). Detecting Performance Anomalies in Digital Advertising Campaigns: A Comparative Evaluation of Rule-Based Monitoring, Statistical Control Charts, and Machine-Learning Detectors. Journal of Innovative Technologies, 2(1).

[17]Mo, Z., & Luo, Z. (2026). A Comparative Evaluation of Rejection Inference Methods for Credit Application Screening: Inverse Probability Weighting, Hard-Cutoff Augmentation, and Semi-Supervised Self-Training. Spectrum of Research, 6(2).

[18]Cao, H., & Yu, M. (2026). Do Neighborhood-Aggregated Features Improve Gradient Boosting for Node-Level Fraud Detection? An Empirical Study on Two Multi-Relational Graphs. Journal of Selected Topics in Academic Research, 2(1).

[19]Lu, Y., & Tang, R. (2026). Comparing Concept-Drift Detection Strategies for Financial Fraud Detection: A Controlled Simulation Study of Detection Latency, False Alarms, and Recovery Across Drift Regimes. Academia Nexus Journal, 5(3).

[20]Wen, S., & Pan, H. (2026). A Comparative Empirical Study of Graph Neural Network Methods for Node-Level Financial Fraud Detection on Public Benchmark Graphs. Journal of Innovative Technologies, 2(1).

[21]Li, Y., & Fu, H. (2026). Interpretable Gradient Boosting for Transmission Capital Expenditure Deviation: A Multi-Decade Benchmark on FERC-Regulated Utilities. Academia Nexus Journal, 5(2).

[22]Lu, Y. (2026, May). Comparative Evaluation of Machine Learning-Based Validation Strategies for Anti-Money Laundering Compliance Data Pipelines. In Proceedings of the 2026 3rd Guangdong-Hong Kong-Macao Greater Bay Area International Conference on Digital Economy and Artificial Intelligence (pp. 1164-1172).

[23]Zheng, Y., Wang, J., & Zhang, X. (2026). Error Patterns of Large Language Model Agents in Multi-Step Web Navigation: A Comparative Empirical Study across Prompting Strategies and Backends. Academia Nexus Journal, 5(2).

[24]Zheng, Y., & Tang, R. (2026). Quantifying the Throughput–Tail-Latency Trade-off of Continuous Batching and Chunked-Prefill Scheduling under Production LLM Traces: An Empirical Comparison. Spectrum of Research, 6(2).

[25]Luo, C., Zheng, Y., & Wang, J. (2026). An Empirical Comparison of Popularity-Debiasing Strategies in News and Short-Video Recommendation. Journal of Innovative Technologies, 2(1).

[26]Zheng, Y., Wang, J., & Lai, J. (2026). Benchmarking KV-Cache Management and Chunked-Prefill Scheduling across Mainstream LLM Inference Engines under Production Traces. Journal of Science, Innovation & Social Impact, 2(4), 74-84.

[27]Zhang, X., Zheng, Y., & Hao, C. (2026). Comparative Empirical Evaluation of Prompting Strategies for Large-Language-Model Web-Navigation Agents on the WebArena Benchmark. Journal of Science, Innovation & Social Impact, 2(4), 25-36.

[28]Yang, H., & Finkelstein, J. (2025, August). Unveiling Pharmacogenomic Patterns in Breast Cancer Through Big Data. In Medinfo 2025-Healthcare Smart x Medicine Deep: Proceedings of the 20th World Congress on Medical and Health Informatics (pp. 1079-1083). 1 Oliver's Yard, 55 City Road, London, EC1Y 1SP: SAGE Publications.

[29]Yang, H., & Finkelstein, J. (2025). Using Big Data to Uncover Drug-Gene Interaction in Patients with Prostate Cancer. In Envisioning the Future of Health Informatics and Digital Health (pp. 131-135). 1 Oliver's Yard, 55 City Road, London, EC1Y 1SP: SAGE Publications.

[30]Yu, M., & Li, Z. (2025). An Empirical Comparison of Discrete Video Tokenization Schemes for Video Question Answering and Video Captioning. Journal of Artificial Intelligence Review, 6(2), 27-50.

[31]Wen, S. (2025). Cross-lingual Robustness of Behavioral Coordination Signals for Detecting State-Linked Information Operations on Twitter: A Comparative Empirical Study. Journal of Artificial Intelligence Review, 6(3), 46-57.

[32]Zheng, Y., Wang, J., & Lai, J. (2026). Benchmarking KV-Cache Management and Chunked-Prefill Scheduling across Mainstream LLM Inference Engines under Production Traces. Journal of Science, Innovation & Social Impact, 2(4), 74-84.

[33]Deng, M., Fu, H., & Sun, Y. (2026). Incremental Contribution of Linguistic, Behavioral, and Propagation Signals to Early Detection of LLM-Generated Coordinated Inauthentic Campaigns. Spectrum of Research, 6(1).

[34]Tian, Y., Pan, H., & Sun, Y. (2026). A Comparative Network-Centrality, Community-Detection, and Cascade-Simulation Analysis of Disruption Propagation and Concentrated Vulnerability in a Real Inter-Sectoral Supply Network. Journal of Selected Topics in Academic Research, 2(1).

[35]Hao, C., & Tian, Y. (2026). Do Maintenance and Failure–Aging Features Improve Probabilistic Forecasting Calibration and Inventory Value for Aerospace Spare Parts? An Empirical Comparison. Journal of Innovative Technologies, 2(1).

[36]Wu, Z., & Chen, J. (2026). A Controlled Comparison of Data-Augmentation Strategies for Surface-EMG Hand-Gesture Recognition under Small-Sample Conditions. Journal of Selected Topics in Academic Research, 2(1).

[37]Luo, C., Zheng, Y., & Wang, J. (2026). An Empirical Comparison of Popularity-Debiasing Strategies in News and Short-Video Recommendation. Journal of Innovative Technologies, 2(1).

[38]Zheng, Y., Wang, J., & Zhang, X. (2026). Error Patterns of Large Language Model Agents in Multi-Step Web Navigation: A Comparative Empirical Study across Prompting Strategies and Backends. Academia Nexus Journal, 5(2).

[39]Chen, Y. (2026). Comparing prompting strategies and backbone choices for LLM agents on tabular credit-card fraud detection: An empirical study. Journal of Innovative Technologies, 2(1).

[40]Wen, S., & Pan, H. (2026). How Far Have We Debiased? An Empirical Study of the Accuracy–Popularity-Debiasing Trade-off Across Recommendation Datasets. Journal of Innovative Technologies, 2(2).

[41]Wang, H., & Sun, Y. (2026). An Empirical Analysis of the Temperature Hyperparameter in Knowledge Distillation for Image Classification Student Networks. Journal of Innovative Technologies, 2(1).

[42]Wang, H. (2026). Position Bias and Verbosity Bias in LLM-as-a-Judge: An Empirical Cross-Benchmark Comparison. Spectrum of Research, 6(1).

[43]Peng, Y., & Gao, S. (2026). Quantifying Risk-Factor Attribution in Utility-Scale Power-Project Returns: A SHAP-Based Comparative Study of Ensemble Learners for Unlevered-IRR Distribution Prediction. Academia Nexus Journal, 5(2).

[44]Zheng, Y., & Tang, R. (2026). Quantifying the Throughput–Tail-Latency Trade-off of Continuous Batching and Chunked-Prefill Scheduling under Production LLM Traces: An Empirical Comparison. Spectrum of Research, 6(2).

[45]Fu, H., & Tang, R. (2026). A Comparative Evaluation of Popularity Debiasing Methods for Mitigating Feedback Loops in Recommendation. Annals of Applied Sciences, 7(1).

[46]Sun, Y., & Tang, R. (2026). A Comparative Evaluation of Post-hoc Explainability Methods for Graph Neural Networks in Node Classification. Journal of Innovative Technologies, 2(2).

[47]Tang, R., & Zhao, F. (2026). An Empirical Comparison of Heuristic and Neural Session-Based Recommendation Methods under Sparse Click-Stream Cold-Start Conditions. Academia Nexus Journal, 5(2).

[48]Peng, Y., & Gao, S. (2026). Does Exit-Timing Flexibility Pay? An Empirical Least-Squares Monte Carlo Assessment of Equity Divestment Timing for Utility-Scale Solar in the PJM Market. Journal of Innovative Technologies, 2(1).

[49]Wang, J., & Guo, Y. (2026). Road-Geometry, Scenario-Diversity, and Failure-History Prioritization for BeamNG Lane-Keeping Simulation Regression Tests under Fixed Simulation-Time Budgets: An Empirical Comparison. Spectrum of Research, 6(2).

[50]Zhang, X., & Zhang, Y. (2026). Statistical, Tree-Ensemble, and Neural Baselines for Electricity Customer Behavior Analysis: A Comparative Evaluation across Load-Shape Segmentation, Multi-Horizon Forecasting, and Theft Detection. Journal of Selected Topics in Academic Research, 2(1).

[51]An Empirical Comparison of Representation-Learning Clustering and Causal Effect Estimation for Electricity Customer Segmentation and Demand Response Targeting

[52]Li, J. (2026, March). Domain-Adaptive Annotation Techniques for Multilingual Cultural Heritage Documents: A Few-Shot Learning Approach. In Proceedings of the 2026 5th International Conference on Cyber Security, Artificial Intelligence and Digital Economy (pp. 667-676).

[53]Luo, Z. (2026, March). An Adaptive Ensemble Learning Approach with Cost-Sensitive Optimization for Credit Default Prediction. In Proceedings of the 2026 5th International Conference on Cyber Security, Artificial Intelligence and Digital Economy (pp. 523-531).

[54]Tian, Y. (2026, March). A Comparative Study of Machine Learning Algorithms for Supplier Delivery Time Prediction in Aerospace Supply Chains: Feature Analysis and Risk Assessment Enhancement. In Proceedings of the 2026 5th International Conference on Cyber Security, Artificial Intelligence and Digital Economy (pp. 532-539).

[55]Zhang, Y. (2026, March). Semantic Prior Embedding for Language-Guided Medical Image Reconstruction Quality Enhancement. In Proceedings of the 2026 6th International Conference on Bioinformatics and Intelligent Computing (pp. 910-915).

[56]Wu, Z. (2026, May). Quality Assessment Framework for Heterogeneous Sensor Data Standardization in Multimodal Research. In Proceedings of the 2026 6th International Conference on Internet of Things and Machine Learning (pp. 672-681).

[57]Chen, X. (2026, May). AI-Enhanced Experimental Evaluation of Cross-Channel Attribution Strategies for Digital Advertising Optimization. In Proceedings of the 2026 3rd Guangdong-Hong Kong-Macao Greater Bay Area International Conference on Digital Economy and Artificial Intelligence (pp. 414-423).

[58]Weng, H., Yang, Q., Fu, H., Pan, H., & Lv, X. (2026). When Retrieval Hurts Code Completion: A Diagnostic Study of Stale Repository Context. arXiv preprint arXiv:2605.14478.

[59]Tian, Y., Pan, H., & Sun, Y. (2026). A Comparative Network-Centrality, Community-Detection, and Cascade-Simulation Analysis of Disruption Propagation and Concentrated Vulnerability in a Real Inter-Sectoral Supply Network. Journal of Selected Topics in Academic Research, 2(1).

[60]Pan, H., Sun, Y., & Wu, Y. (2026). Can Task-Prefixed Text Embeddings Exploit Anonymized Financial-Graph Features? A Low-Label Comparison on T-Finance and Elliptic. Academia Nexus Journal, 5(3).

[61]Tang, R., Pan, H., & Yang, C. (2026). Effects of Prompt Example Count on Money-Laundering Transaction Classification from Textualized Financial-Graph Neighborhoods: An Empirical Study. Spectrum of Research, 6(2).

[62]Wen, S., & Pan, H. (2026). How Far Have We Debiased? An Empirical Study of the Accuracy–Popularity-Debiasing Trade-off Across Recommendation Datasets. Journal of Innovative Technologies, 2(2).

[63]Pan, H., & Zhong, M. (2026). Concept-Drift Detection for Systemic-Risk Early Warning: A Comparative Evaluation of Regime-Conditional Adaptive Thresholding under Distribution Shifts in Financial Stress Indices. Annals of Applied Sciences, 7(1).

[64]Chen, X., Fu, H., & Pan, H. (2026). Detecting Performance Anomalies in Digital Advertising Campaigns: A Comparative Evaluation of Rule-Based Monitoring, Statistical Control Charts, and Machine-Learning Detectors. Journal of Innovative Technologies, 2(1).

[65]Wen, S., & Pan, H. (2026). A Comparative Empirical Study of Graph Neural Network Methods for Node-Level Financial Fraud Detection on Public Benchmark Graphs. Journal of Innovative Technologies, 2(1).

[66]Zhang, Y., Yang, C., Ke, H., Chan, K. C., & Wang, W. (2024). A study on the microstructure and mechanical behavior of CoCrFeNi high entropy alloy fabricated via laser powder bed fusion: Experiment and crystal plasticity finite element modelling. Materials Science and Engineering: A, 893, 146111.

[67]Yang, C., Ding, J., Qu, S., Ouyang, D., Zhang, L., Zhang, Y., ... & Wang, W. H. (2025). High strength bioinspired cellular metallic glasses with excellent energy absorption. Acta Materialia, 285, 120688.

[68]Chen, C., Pu, Z., Qin, B., Wang, P., Yang, C., & Chan, K. (2024). Elastocaloric effect of NiTi shape memory alloys manufactured by laser powder bed fusion. Journal of Materials Research and Technology, 33, 3439-3451.

[69]Yang, C., Ouyang, D., Zhang, L., Zhang, Y., Tong, X., Ke, H., ... & Wang, W. (2024). The enhancement of damage tolerance of 3D-printed high strength architected metallic glasses by unit cell shape design. Additive Manufacturing, 85, 104125.

[70]Zhang, Y. Y., Yang, C. R., Tong, X., Zhou, J., Liu, L., Xiao, M., ... & Wang, W. H. (2025). Oxidation sequence modulation induced superior high-temperature tribological performance of Ti-Hf-Nb-V refractory high entropy alloy fabricated through directed energy deposition: Y.-Y. Zhang et al. Rare Metals, 44(4), 2695-2704.

[71]Yang, C., Pei, C., Ke, H., Sun, S., Zhang, H., Ouyang, D., ... & Wang, W. H. (2026). Topology-engineered metallic glass metamaterial-supported NiMo nanoglass catalysts for ultra-efficient water splitting. National Science Review, nwag442.

[72]Zhang, Y., Yu, K., Qin, B., Yang, C., Ye, S., Feng, C., ... & Wang, W. (2024). Origins of strength stabilities at elevated temperatures in additively manufactured refractory high entropy alloy. Materials Science and Engineering: A, 915, 147225.

[73]Yang, C., Ding, J., Qu, S., Ouyang, D., Zhang, L., Zhang, Y., ... & Wang, W. High Strength Bioinspired TPMS-Architected Metallic Glasses with Excellent Energy Absorption. Available at SSRN 4981554.

[74]Chen, C. Y., Pu, Z., Chen, W., Qin, B. L., Luo, H., Wang, P. B., ... & Chan, K. C. (2026). Inhomogeneity of Room-Temperature Elastocaloric Effects in NiTi Alloys Fabricated by Laser Powder Bed Fusion. Rare Metals, 45(1), e70044.

Budgeted Human Review for Privacy-Aware Clinical and Financial Risk Triage