Self-Optimizing Integration Pipelines for Dynamic Clinical Workloads: A Reinforcement Learning Approach to Autonomous Performance Tuning in Healthcare Middleware
DOI:
https://doi.org/10.63282/3050-9262.IJAIDSML-V7I3P114Keywords:
Reinforcement Learning, Self-Optimization, Healthcare Middleware, Dynamic Workloads, Autonomous Tuning, Integration Engine, Performance Optimization, Adaptive Systems, Hl7, FHIR, PPOAbstract
Healthcare integration engines operate under dynamic workload conditions that vary across diurnal cycles, weekly patterns, seasonal census fluctuations, and event-driven surges. Optimal engine performance depends on numerous configurable parameters including thread pool sizes, polling intervals, batch processing thresholds, queue capacity limits, connection pool allocations, and timeout durations that must be tuned to match current workload characteristics. In practice, these parameters are statically configured and rarely adjusted, leading to chronic suboptimal performance: over-provisioned during low-volume periods and under-provisioned during peak demand. This paper introduces the Adaptive Pipeline Optimization Agent (APOA) a reinforcement learning system using Proximal Policy Optimization (PPO) that continuously observes integration engine telemetry and autonomously adjusts pipeline configuration parameters to optimize throughput, minimize latency, and prevent resource exhaustion under dynamically changing conditions. Evaluation across a simulated enterprise environment 180 channels, 500 messages/second, four distinct workload regimes demonstrates that APOA achieves 31% improvement in peak-hour throughput, 44% reduction in 95th-percentile processing latency, and 18% better resource utilization efficiency compared to static configuration and rule-based autoscaling baselines. APOA adapts to a previously unseen workload pattern within 4.2 minutes, compared to 22 minutes for manual operator response. All targets for zero message loss and ordering preservation are met.
References
[1] R. Haux, "Health information systems — past, present, future," Int. J. Med. Inform., vol. 75, pp. 268–281, 2006.
[2] D. Bender and K. Sartipi, "HL7 FHIR: An agile and RESTful approach to healthcare information exchange," in Proc. IEEE CBMS, 2013, pp. 326–331.
[3] D. W. Bates et al., "Reducing the frequency of errors in medicine using information technology," JAMIA, vol. 8, no. 4, pp. 299–308, 2001.
[4] S. Sundaram, "A fault-tolerant timeout framework for external service calls in healthcare integration engines," The American J. Eng. Technol., vol. 8, no. 3, pp. 121–126, 2026, doi: 10.37547/tajet/v8i3-323.
[5] Sundaram, S. (2026). Predictive Failure Detection in Healthcare Integration Middleware Using Hybrid Ensemble Time-Series Machine Learning. International Journal of Artificial Intelligence, Data Science, and Machine Learning, 7(2), 27-35. https://doi.org/10.63282/3050-9262.IJAIDSML-V7I2P105
[6] J. O. Kephart and D. M. Chess, "The vision of autonomic computing," Computer, vol. 36, no. 1, pp. 41–50, 2003.
[7] T. Lorido-Botran, J. Miguel-Alonso, and J. A. Lozano, "A review of auto-scaling techniques for elastic applications in cloud environments," J. Grid Computing, vol. 12, no. 4, pp. 559–592, 2014.
[8] D. Van Aken et al., "Automatic database management system tuning through large-scale machine learning," in Proc. ACM SIGMOD, 2017, pp. 1009–1024.
[9] Z. Xu et al., "Experience-driven networking: A deep reinforcement learning based approach," in Proc. IEEE INFOCOM, 2018, pp. 1871–1879.
[10] HL7 International, "HL7 v2.x Messaging Standard," 2019. [Online]. Available: hl7.org/imple...d=185
[11] R. S. Sutton and A. G. Barto, Reinforcement Learning: An Introduction, 2nd ed. MIT Press, 2018.
[12] J. Schulman, F. Wolski, P. Dhariwal, A. Radford, and O. Klimov, "Proximal policy optimization algorithms," arXiv preprint arXiv:1707.06347, 2017.
[13] H. Mao, M. Alizadeh, I. Menache, and S. Kandula, "Resource management with deep reinforcement learning," in Proc. ACM HotNets, 2016, pp. 50–56.
[14] M. Nygard, Release It! Design and Deploy Production-Ready Software, 2nd ed. Pragmatic Bookshelf, 2018.
[15] A. R. Hevner et al., "Design science in information systems research," MIS Quarterly, vol. 28, no. 1, pp. 75–105, 2004.
[16] J. Walonoski et al., "Synthea: An approach, method, and software mechanism for generating synthetic patients and the synthetic electronic health care record," JAMIA, vol. 25, no. 3, pp. 230–238, 2018.
