A CARL Framework for Scalable Processing of Massive Distributed Electronic Health Record (EHR) Streams in Multi-Cloud Environments

• Billy Elly

  Lautech

  Author

The proliferation of electronic health record (EHR) systems, coupled with the emergence of multi-cloud healthcare architectures, has created unprecedented challenges in processing massive, heterogeneous clinical data streams while maintaining cost efficiency and service-level agreement (SLA) compliance. Existing approaches to EHR stream processing predominantly rely on static resource provisioning or rule-based auto-scaling mechanisms that fail to adapt to the stochastic nature of clinical workloads and the cost variability inherent in multi-cloud environments. This study proposes and validates a Cost-Aware Reinforcement Learning (CARL) framework that models multi-cloud resource allocation for EHR stream processing as a sequential decision-making problem, dynamically balancing computational performance against operational costs. Using a design-based research methodology combining retrospective analysis of synthetic EHR workloads and prospective simulation across three simulated cloud providers, the framework was evaluated against baseline auto-scaling approaches. Results demonstrate that CARL achieved 89.4% SLA compliance while reducing operational costs by 31.7% compared to static provisioning methods, with statistically significant improvements in resource utilization efficiency (p < 0.001). Feature importance analysis identified workload variance, inter-cloud latency differentials, and spot instance availability as the strongest predictors of optimal scaling decisions. The study contributes a replicable framework for cost-aware EHR stream processing that extends reinforcement learning applications in healthcare informatics, offering practical guidance for healthcare system administrators navigating the complex trade-offs between clinical data processing requirements and cloud infrastructure expenditure.

• pdf

Copyright (c) 2026 Billy Elly (Author)

This work is licensed under a Creative Commons Attribution 4.0 International License.