As health systems continue to face escalating pressures, from labor shortages and capacity constraints to continued margin compression, leaders are increasingly exploring how artificial intelligence (AI) can strengthen utilization management (UM), enhance clinical alignment, and improve operational flow. Yet many executives remain cautious, asking a critical question: Where is the evidence that AI improves outcomes in real-world settings?
A growing body of peer-reviewed research now offers compelling confirmation. Across three major health systems — Baylor Scott & White, Yale New Haven Health, and Mayo Clinic Health System — AI-driven predictive models have demonstrated measurable improvements in accuracy, resource forecasting, level-of-care determination, and reimbursement integrity. These independent evaluations published in academic journals provide a clearer picture of what health systems can and should expect when integrating AI into core operational workflows.
The Challenge: Persistent Variability and Capacity Strain in Concurrent Authorization
Decades of research show that utilization management remains highly variable across organizations. Studies have documented that level-of-care determinations often differ by reviewer, by shift, or by incomplete capture of comorbid conditions. Meanwhile, observation rates and length-of-stay pressures continue to grow nationally, contributing to bottlenecks and lost revenue opportunities.
Simultaneously, many hospitals struggle to forecast resource needs in real time. Misaligned discharge planning and inaccurate LOS expectations reduce throughput efficiency and add to avoidable delays. These systemic challenges have created an environment where even incremental improvements can produce large downstream impacts.
AI and the Alignment of Level of Care: Evidence From Baylor Scott & White
One of the most extensive evaluations comes from Baylor Scott & White Health, where researchers examined the accuracy of an AI model, the Care Level Score™ (CLS), designed to support early level-of-care predictions. According to a Baylor University Medical Center Proceedings study1, the model correctly predicted inpatient discharge status in 86 percent of cases when it indicated an inpatient level of care based on the clinical merit of the case.
The implications for healthcare leaders are significant. Early, objective predictions, often within the first 12 hours, allow physicians, case managers, and utilization review teams to align level of care decisions more consistently. This alignment not only reduces administrative friction but also supports right-level placement for patients while protecting appropriate reimbursement.
Reducing Unnecessary Observation Stays: Insights From Yale
In a Journal of Doctoral Nursing Practice study2 featuring Yale New Haven Health, researchers studied the impact of integrating AI into UM workflows to strengthen clinical appropriateness for inpatient designation. The results showed a reduction in observation discharge rates from 16.69% to 12.75%, signaling more precise identification of patients who require an inpatient status based on clinical documentation.
Nurses reported that AI-supported insights helped them surface relevant comorbidities, engage in more informed discussions with providers, and advocate more effectively for clinically appropriate designations. For leaders navigating a landscape where observation status remains a costly and contentious issue, the study findings suggest that AI can serve as a stabilizing force, reducing variation, supporting compliant decisions, and improving reimbursement through better alignment.
Forecasting Throughput and Resource Needs: Mayo’s GMLOS Validation
A recent Journal of Clinical and Translational Science study3 featuring Mayo Clinic expanded the lens beyond concurrent authorization, evaluating AI’s ability to forecast DRGs and predict geometric mean length of stay (GMLOS). Their findings showed 81% accuracy for the top three DRG predictions and, notably, a near-negligible variance of ±0.14 days between predicted and actual GMLOS.
Hospital leaders increasingly highlight length of stay (LOS) predictability as a critical operational lever. When clinicians trust LOS projections, they can more effectively coordinate multidisciplinary rounds, anticipate bed needs, and streamline discharge planning. Mayo clinicians found the GMLOS prediction as the most valuable output, because it directly improved their ability to manage flow and throughput.
What This Means for Health System Strategy
Across these independent studies, it is illustrated that utilizing AI in concurrent authorization:
- Improves accuracy, reducing variation in level-of-care determinations.
- Increases operational efficiency, with measurable reductions in unnecessary observation stays and more informed discharge planning.
- Enables more precise resource forecasting, strengthening capacity management and throughput.
- Provides clinical teams with decision support, enhancing confidence rather than circumventing clinical judgment.
For healthcare leaders evaluating AI investment, these studies underscore that the potential benefits are no longer theoretical. Objective, published evidence now demonstrates that AI can directly influence operational outcomes and financial performance when thoughtfully integrated into existing workflows.
Read more about the three peer-reviewed studies here– https://www.xsolis.com/peer-reviewed-studies/.
[1] Watson, L. E., Light, R. A., & Shaver, C. (2025). Ability of artificial intelligence to correctly predict inpatient versus observation hospital discharge status. Baylor University Medical Center Proceedings, 38(5), 662–665. https://pmc.ncbi.nlm.nih.gov/articles/PMC12351743/
[2] Tuccio L, Catapano T, Elwell J, Dupont N, Sines E, Pisanelli F Jr. Leveraging Artificial Intelligence to Improve Clinical Appropriateness of Inpatient Designation in a Utilization Management Setting. J Dr Nurs Pract. 2025 Sep 29:JDNP-2025-0034.R1. doi: 10.1891/JDNP-2025-0034. Epub ahead of print. PMID: 41022622. https://pubmed.ncbi.nlm.nih.gov/41022622/
[3] Muhanga A, Jason K, Mueller X, et al. 355 Validation of an artificial intelligence Algorithm for predicting diagnosis-related groups in a community health system. Journal of Clinical and Translational Science. 2025;9(s1):109-109. doi:10.1017/cts.2024.981. https://pmc.ncbi.nlm.nih.gov/articles/PMC12050694/
