Background
A significant challenge in utilizing continuous patient monitor data for AI/ML predictions of catastrophic clinical events is deploying scalable platforms that can ingest, normalize, and deliver patient-centric measurements to algorithm modules in real-time. We describe a stream processing platform capable of simultaneously ingesting and normalizing data from multiple hospitals.
• Secure cloud-based AWS environment
• Multi-modal patient data processing from various sources (PMs and EMRs)
• Real-time data transformation and management
• Configurable live data feed system
• Standalone software module support (Python functions)
• Stream processing engine based on Apache Kafka
Our platform successfully deployed three applications:
• NEWS score implementation (NEWS-vital)
• Heart rate variability prediction using logistic regression models
• In-hospital cardiac arrest prediction using deep neural networks (transformer architecture)
The Digital Health Platform (DHP) processes data through:
• Two-minute ECG waveform windows for HRV analysis
• Time series features including:
o Heart rate dynamics from ECG waveforms
o Statistical features from vital signs
o Cross-correlation between 30-minute vital-sign time series
Testing methodology:
• Streamed clinical data from 100 patient monitors
• Used data playback engine with 10 repetitions
• Evaluated CPU, memory, disk, and bandwidth usage
Our platform demonstrated:
• Rapid and repeatable algorithm deployment
• Consistent performance across varying complexities
• Successful real-time result display through web-based viewing
• Verified processing capacity for hospital-wide deployment
Live hospital-wide deployment of algorithms requires:
• Scalable platform for bedside monitor data and ECG waveforms
• Windowed data delivery to analytical modules
• Efficient distribution of results to clinical applications
Our research demonstrates that:
• Stream processing engine based on Apache Kafka
• Integration with Pyspark
• Enables rapid algorithm deployment regardless of complexity
• Timothy Ruchti, PhD
• Abel Lin, BS
• Mohamed Elmahdy, PhD
• Elias Bitar, MBA MSE
• Jessa Deckwa, BS
Interested in implementing this platform at your facility?
Contact our team to discuss next steps:
Elias Bitar
elias_bitar@nihonkohden.com
