Generating Real‐World Evidence From AI Deployed in Health Systems
8 affiliations
- 1Rady School of Management University of California San Diego California USA
- 2Joan and Irwin Jacobs Center for Health Innovation UC San Diego Health San Diego California USA
- 3Division of Biomedical Informatics, Department of Medicine University of California San Diego California USA
- 4Division of General Internal Medicine, Department of Medicine University of California San Diego California USA
- 5Department of Electrical and Computer Engineering University of California San Diego California USA
- 6Department of Biology University of California San Diego California USA
- 7Division of Cardiovascular Medicine, Department of Medicine University of California San Diego California USA
- 8Department of Neurological Surgery University of California San Diego California USA
Abstract
Background: Health systems increasingly deploy predictive and generative artificial intelligence (AI) to improve care delivery and outcomes. Predictive AI models estimate the risk of future events to inform targeted interventions, while generative AI systems extract, summarize, and communicate information and may automate selected tasks. Routine deployment generates clinical, operational, and interaction data that can support evaluation of the effectiveness and safety of AI‐enabled workflows.
Methods: We review approaches to generating real‐world evidence (RWE) from routine deployment of predictive and generative AI across four implementation configurations: silent deployment, human‐in‐the‐loop (HITL), human‐on‐the‐loop (HOTL), and automated. For each configuration, we describe the evaluation questions, routinely generated data, and feasible study designs.
Results: Each implementation configuration supports different forms of evidence about model performance, workflow effectiveness, human oversight, and safety. For both predictive and generative AI, silent deployment supports assessment of local performance and subgroup differences. Once predictive AI drives action, evaluating the effectiveness of an implemented workflow (HITL, HOTL, or automated) generally requires causal inference approaches. For generative AI, HITL implementations can generate data on edits, overrides, time on task, usability, cognitive burden, and safety‐relevant failure modes. HOTL and automated generative AI implementations can generate data on escalation performance, human follow‐up work, downstream outcomes, and safety signals.
Conclusion: AI implementation should be paired with methods for generating RWE on the effectiveness and safety of AI‐enabled workflows. In HOTL and automated configurations, escalation performance and residual human workload become particularly important components of evaluation.
Related
- Greater and Clinically Meaningful Improvement in Health-Related Quality of Life After Device-Aided Therapy in Younger Patients Earlier Within Advanced Parkinson’s Disease: Real-World Evidence from the DATs-PD GETM Spanish Registry
- Integrating Real-World Evidence, Patient-Reported Outcomes and Health Economics into Evidence-Based Public Health-A Narrative Review and Conceptual Framework.
- TRENDS IN HOSPITALIZATIONS, COSTS, AND IN-HOSPITAL MORTALITY IN PRIMARY BILIARY CHOLANGITIS: A DECADE OF REAL-WORLD EVIDENCE FROM BRAZIL’S UNIFIED HEALTH SYSTEM Tendências nas hospitalizações, custos e mortalidade intra-hospitalar por colangite biliar primária: uma década de evidências de mundo real do Sistema Único de Saúde do Brasil
- Exploratory Mendelian Randomization Provides Hypothesis-Generating Evidence for a Possible Nicotine-Related Mechanism in Sarcoidosis
- Patterns and clinical outcomes of antidepressant prescribing in children and adolescents: real-world evidence from a mental health hospital in Türkiye
BibTeX
@article{poursoltan2026generating,
title = {{Generating Real‐World Evidence From AI Deployed in Health Systems}},
author = {Lily Poursoltan and Aaron Boussina and Jie Cao and Eileen Kim and Poorva Satish Bedmutha and Hetvi Majmudar and Mattheus Ramsis and Alexander Khalessi and Karandeep Singh},
journal = {Learning Health Systems},
year = {2026},
month = nov,
doi = {10.1002/lrh2.70131},
url = {https://doi.org/10.1002/lrh2.70131},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Generating Real‐World Evidence From AI Deployed in Health Systems
AU - Lily Poursoltan
AU - Aaron Boussina
AU - Jie Cao
AU - Eileen Kim
AU - Poorva Satish Bedmutha
AU - Hetvi Majmudar
AU - Mattheus Ramsis
AU - Alexander Khalessi
AU - Karandeep Singh
PY - 2026
DA - 2026/11/04
JO - Learning Health Systems
DO - 10.1002/lrh2.70131
UR - https://doi.org/10.1002/lrh2.70131
AB - Background: Health systems increasingly deploy predictive and generative artificial intelligence (AI) to improve care delivery and outcomes. Predictive AI models estimate the risk of future events to inform targeted interventions, while generative AI systems extract, summarize, and communicate information and may automate selected tasks. Routine deployment generates clinical, operational, and interaction data that can support evaluation of the effectiveness and safety of AI‐enabled workflows. Methods: We review approaches to generating real‐world evidence (RWE) from routine deployment of predictive and generative AI across four implementation configurations: silent deployment, human‐in‐the‐loop (HITL), human‐on‐the‐loop (HOTL), and automated. For each configuration, we describe the evaluation questions, routinely generated data, and feasible study designs. Results: Each implementation configuration supports different forms of evidence about model performance, workflow effectiveness, human oversight, and safety. For both predictive and generative AI, silent deployment supports assessment of local performance and subgroup differences. Once predictive AI drives action, evaluating the effectiveness of an implemented workflow (HITL, HOTL, or automated) generally requires causal inference approaches. For generative AI, HITL implementations can generate data on edits, overrides, time on task, usability, cognitive burden, and safety‐relevant failure modes. HOTL and automated generative AI implementations can generate data on escalation performance, human follow‐up work, downstream outcomes, and safety signals. Conclusion: AI implementation should be paired with methods for generating RWE on the effectiveness and safety of AI‐enabled workflows. In HOTL and automated configurations, escalation performance and residual human workload become particularly important components of evaluation.
SN - 2379-6146
ER -