High-Frequency Data Ingestion and Analytics

Problem Definition

Traditionally, patient data obtained from health monitoring devices (BP monitoring, oxygen sensors, ECG/EKG, etc.) in the inpatient setting is stored in Electronic Health Records (EHRs) at a much lower frequency than the devices are generating. Consequently, some of the important data points that may predict adverse events or indicate issues are potentially being lost. There is a strong interest within healthcare institutions to capture this data at the original source frequency and use this data for advanced predictive analytics, precision medicine, overall care quality and safety improvement, and clinical research. However, capturing and storing this data requires extensive computing power and a drastic increase in storage capacity.

Approach

Clinovera has been engaged by InterSystems, a leading Healthcare technology organization, to develop advanced high-frequency medical data ingestion solutions and workflows that facilitate effective and actionable analytics while minimizing infrastructure and storage requirements. 

InterSystems has defined the following high-level objectives for this project:

• Develop and validate the data model for compressed and efficient storage of device data
• Achieve high-performance metrics in real-time data acquisition from a wide range of devices across the entire hospital
• Design and implement data exposure in a format suitable for efficient analytics
• Develop or integrate Business intelligence (BI) tools over device data storage

Implementation

All medical devices are connected to the Philips Capsule Gateway Server which converts low-level device protocols to HL7 messaging. The data from the Capsule Gateway is processed and sent to Epic EHR at a low frequency. For this project, the HL7 device data feed (waveforms, alerts, and trends) is replicated and aggregated in InterSystems IRIS for Health Advanced Server infrastructure at the source frequency. 

Clinovera has mapped and transformed the HL7 data to the specialized InterSystems data model, Clinovera further refined and improved message processing, optimized storage, and improved performance achieving an 8X compression rate and throughput of over 700K HL7 messages per second. 

The team has also implemented SQL mapping and exposure of compressed storage representation to surface data to analytics tools used by the client team. Utilizing auxiliary indexing and asynchronous precomputed metrics, the team has achieved high-performance data retrieval results. 

The joint InterSystems/Clinovera team continues to work with the client institution to define necessary tools for predictive modeling and analytics to enhance care.