The University of Texas Health Science Center at Houston (UTHealth) is conducting research related to the prevalence of subarachnoid hemorrhage (SAH) in the presence of vasopressor medication. Virtusa collaborated with UTHealth, Cardinal Health, and Amazon Web Services (AWS), to use AI and machine learning techniques to take this medical research forward.
The research aims to find the best treatment and management strategies for SAH and diabetes, based on healthcare data and computer simulations.
UTHealth conducts the research by training multivariate machine learning models on a training dataset provided by a third-party vendor.
Purchasing this data and releasing the data to test hypotheses presented in research publications is an expensive affair and poses high PHI risks for UTHealth.
As a result, they wanted to use simulated medical data that does not present any PHI risk to validate their research and enable safe publication alongside the research for the scientific community.
Virtusa collaborated with Cardinal Health to simulate and provide this data using Cardinal Health’s ProxiTM product which runs on Virtusa’s vLifeTM platform and consists of three main components:
- Synthesizer – It consists of a Java application running on multiple EC2 instances.
- Simulated data lake – The synthesizer writes parquet data files to S3 buckets, which are then exposed through AWS Redshift Spectrum to form the data lake.
- ProxiHub Portal – It consists of a web portal UI that runs on AWS CloudFront and a backend web service API server that runs on AWS Elastic Beanstalk.
Using vLife, Virtusa’s cloud-based platform consisting of a comprehensive HIPAA-compliant data lake with multiple data sources, pre-built APIs, AI, and machine learning models, researchers are trying to discover hidden trends and develop new treatment strategies and cures for a range of illnesses.
- UTHealth is able to leverage simulated clinical records to accelerate research, augment existing real-world-evidence, and save time by using clean ready-to-use linked data.
- The availability of Virtusa’s machine learning model packages for disease state classifications on AWS Marketplace will assist our healthcare data scientist customers in deploying and integrating machine learning models into their applications quickly.
- The data provided through Cardinal Health’s Proxi patient population simulation requires minimum data cleaning and preparation and is ready for the application of machine learning models soon after synthesis.
- Simulated data is delivered to UTHealth in iterations after consolidating the required features of the data, comparing them, and running a simulation. Two simulations of the entire US population have been completed so far, and the data for the SAH cohort has been delivered to UTHealth. UTHealth has deemed the demographics comparisons between their proprietary data and the simulated data to be satisfactory. It means that in terms of patient demographics, the simulated data emulates the data used for the research conducted by UTHealth.