Machine Learning Versus Conventional Benchmarks for Predicting Next-Year Medicare Allowed Amounts
Main Article Content
Abstract
The Medicare Physician Fee Schedule provides an administrative reference for physician payment, but fee-schedule values and historical averages are retrospective and may not capture nonlinear, geographic, specialty-specific, setting-specific, and utilization-related changes that influence future observed allowed amounts. This study evaluates whether machine-learning and conventional models can predict next-year service-weighted average Medicare allowed amounts for high-volume evaluation-and management services more accurately than predetermined fee-schedule and historical benchmarks. Publicly available Centers for Medicare & Medicaid Services (CMS) Medicare Physician and Other Practitioners data and Physician Fee Schedule files from 2018–2024 were used. The analysis was restricted to participating individual physician billing for HCPCS 99213 and 99214 across the 50 U.S. states and the District of Columbia, separated by facility and non-facility setting. Groups required at least 100 services and five unique clinicians. Models were developed on earlier years, selected using 2023 data, and evaluated on a sealed 2024 test set. Histogram gradient boosting (HGB), random forest, ridge regression, pooled-history, recent-year-history, and fee-schedule-reference approaches were compared using service-weighted mean absolute error (WMAE). The final 2024 cohort contained 6,829 eligible groups representing 117,969,671 services. HGB achieved the lowest WMAE at $0.829 per service, compared with random forest at $1.173, ridge regression at $1.650, pooled history at $1.884, recent-year history at $1.907, and the fee-schedule reference at $4.803. Relative to ridge regression, HGB reduced WMAE by 49.7%, with a state-block bootstrap difference of -$0.821 per service (95% CI: -$1.172 to -$0.563). HGB improved WMAE in 49 of 51 jurisdictions, with at least 5% improvement in 47. These findings support temporally sealed, group-level machine-learning forecasting of Medicare allowed amounts for the evaluated services while not establishing claims-level accuracy, causal effects, or future official CMS payment decisions.
Article Details
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.
IJCERT Policy:
The published work presented in this paper is licensed under the Creative Commons Attribution 4.0 International (CC BY 4.0) license. This means that the content of this paper can be shared, copied, and redistributed in any medium or format, as long as the original author is properly attributed. Additionally, any derivative works based on this paper must also be licensed under the same terms. This licensing agreement allows for broad dissemination and use of the work while maintaining the author's rights and recognition.
By submitting this paper to IJCERT, the author(s) agree to these licensing terms and confirm that the work is original and does not infringe on any third-party copyright or intellectual property rights.
References
J. A. Fiore, A. M. Sisko, J. A. Poisal, S. D. Smith, G. A. Cuckler, A. J. Madison, S. P. Keehan, K. E. Rennie, and N. J. Feehley, “National Health Expenditure Projections,2025–34: Strong Utilization Growth Initially, Legislative Impacts Later,” Health Affairs, vol. 45, no. 7, pp. 722–733, 2026, doi: 10.1377/hlthaff.2026.00642.
A. Warrier, S. Iqbal, D. Bu, R. Bartholomew, and D. J. Lee, “Spatiotemporal Medicare Reimbursement Trends and Economic Clusters in Otologic/Neurotologic Surgery,” The Laryngoscope, vol. 136, no. 10, pp. 4444–4456, 2026, doi: 10.1002/lary.70652.
J. Wang, Z. Qin, J. Hsu, and B. Zhou, “A fusion of machine learning algorithms and traditional statistical forecasting models for analyzing American healthcare expenditure,” Healthcare Analytics, vol. 5, Art. no. 100312,2024, doi: 10.1016/j.health.2024.100312.
N. Caravaggio, R. Lagravinese, and G. Resce, “The determinants of health expenditure: a machine learning approach,” Empirical Economics, vol. 70, Art. no. 30,2026, doi: 10.1007/s00181-025-02854-6.
T. S. Dee, “The case for preregistering quasiexperimental program and policy evaluations,” Evaluation Review, vol. 49, no. 5, pp. 931–945, 2025, doi:
1177/0193841X251326738.
D. Lakens, C. Mesquida Caldentey, S. Rasti, and M. Ditroilo, “The benefits of preregistration and Registered Reports,” Evidence-Based Toxicology,
vol. 2, no. 1, Art. no. 2376046, 2024, doi:10.1080/2833373X.2024.2376046.
R. A. Berenson and K. J. Hayes, “The road to value can’t be paved with a broken Medicare Physician Fee Schedule,” Health Affairs, vol. 43, no. 7, pp. 950–958, 2024, doi: 10.1377/hlthaff.2024.00299.
E. W. Christensen, G. N. Nicola, E. Y. Rula, L. P.Nicola, and J. A. Hirsch, “Medicare Volume Growth and Shift in Payments From Physicians to Non-Physician Practitioners Under Statutory Budget Neutrality,” Inquiry,vol. 61, Art. no. 00469580241249076, 2024, doi:10.1177/00469580241249076.
A. Zink, M. E. Chernew, and H. T. Neprash, “How Should Medicare Pay for Artificial Intelligence?” JAMA Internal Medicine, vol. 184, no. 8, pp. 863–864, 2024,doi: 10.1001/jamainternmed.2024.1648.
I. H. Patel, A. Leyva, and M. K. K. Niazi, “A Multi-Schedule Machine Learning Pipeline for Medicare Reimbursement Change Prediction and Operational Risk Stratification,” medRxiv, preprint, 2026, doi: 10.64898/2026.07.22.26358736.
A. Rajaram, M. Judd, and D. Barber, “Deep Learning Models to Predict Diagnostic and Billing Codes Following Visits to a Family Medicine Practice: Development and Validation Study,” JMIR AI, vol. 4, Art. no. e64279, 2025, doi: 10.2196/64279.
K. A. Mani, S. N. Goldman, T. Scharfenberger, V. Shankar, M. Bhatta, R. De la Garza Ramos, M. S. Fourman, and A. S. Eleswarapu, “Multimodal machine learning for risk-stratified bundled payments in spinal surgery,”npj Digital Medicine, vol. 8, Art. no. 512, 2025, doi: 10.1038/s41746-025-01915-5.
B. Langenberger, T. Schulte, and O. Groene, “The application of machine learning to predict high-cost patients: A performance-comparison of different models using healthcare claims data,” PLOS ONE, vol. 18, no. 1, Art.no. e0279540, 2023, doi: 10.1371/journal.pone.0279540.
K. Lopez, H. Li, H. Paek, B. Williams, B. Nath, E. R. Melnick, and A. J. Loza, “Predicting physician departure with machine learning on EHR use patterns: A longitudinal cohort from a large multi-specialty ambulatory practice,” PLOS ONE, vol. 18, no. 2, Art. no. e0280251, 2023, doi: 10.1371/journal.pone.0280251.
T. Dong et al., “Performance Drift in Machine Learning Models for Cardiac Surgery Risk Prediction: Retrospective Analysis,” JMIRx Med, vol. 5, Art. no. e45973, 2024, doi: 10.2196/45973.
P. C. Silva et al., “Longitudinal Model Shifts of Machine Learning-Based Clinical Risk Prediction Models: Evaluation Study of Multiple Use Cases Across Different Hospitals,” Journal of Medical Internet Research, vol.26, Art. no. e51409, 2024, doi: 10.2196/51409.
O. Efthimiou, M. Seo, K. Chalkou, T. Debray, M. Egger,and G. Salanti, “Developing clinical prediction models:a step-by-step guide,” BMJ, vol. 386, Art. no. e078276,2024, doi: 10.1136/bmj-2023-078276.
A. R. Rao, R. Jain, M. Singh, and R. Garg, “Predictive interpretable analytics models for forecasting healthcare costs using open healthcare data,” Healthcare Analytics, vol. 6, Art. no. 100351, 2024, doi: 10.1016/j.health.2024.100351.
E. Brati, A. Braimllari, and A. Gjeçi, “Machine Learning Applications for Predicting High-Cost Claims Using Insurance Data,” Data, vol. 10, no. 6, Art. no. 90, 2025, doi: 10.3390/data10060090.
G. S. Collins et al., “Evaluation of clinical prediction models (part 1): from development to external validation,” BMJ, vol. 384, Art. no. e074819, 2024, doi: 10.1136/bmj-2023-074819.
G. R. Ramya, M. V. Krishna, S. T. Nori, K. P. Yeshwanth, and G. K. Jagannadh, “Interpretable Healthcare Cost Prediction Using Explainable XGBoost with SHAP and LIME,” in Sixth Congress on Intelligent Systems, pp. 47–61, 2026, doi: 10.1007/978-3-032-18135-0_5.
A. Al Ataby, F. Moutassem, and W. Al Nuaimy, “Insurance Cost Prediction with Stacked Ensembles, Conformal Intervals, SHAP Explainability, and Fairness Auditing,” Engineering, Technology & Applied Science Research, vol. 16, no. 4, pp. 38276–38286, 2026, doi:10.48084/etasr.19923.
M. Schaekermann et al., “Health equity assessment of machine learning performance (HEAL): a framework and dermatology AI model case study,” eClinicalMedicine, vol. 70, Art. no. 102479, 2024, doi: 10.1016/j.eclinm.2024.102479.