CHALLENGE
Symphony leverages their patient data in partnership with DataRobot to predict patient readmission risk in order to ultimately reduce readmission rates.
However, utilizing DataRobot's multi-tenant SaaS option was a challenge due to HIPAA regulations for Protected Health Information (PHI). Long legal reviews and manual data transformations caused significant delays, taking over three months to upload a single data set to DataRobot's multi-tenant SaaS cloud.
This bottleneck made retraining the readmission risk model a daunting task and stifled the pursuit of new, value-add machine learning projects.
APPROACH
Subsalt's generative database enables Symphony to programmatically create and manage fully synthetic, de-identified data on demand and at scale.
With a source of granular de-identified data that can power AI/ML projects, Symphony eliminated their privacy and security constraints.
With the bottleneck eliminated, many more ML projects can be pursued without legal reviews or manual data transformations and without compromising on data quality or privacy.
RESULTS
95% Collapse in Time to Data
Time to access Symphony's sensitive data was reduced from approximately 3 months to less than a day.
Real Performance, Synthetic Data
Performance of predictive models was maintained using synthetic data as compared to models trained on sensitive source data.
All the benefits, none of the risk.
AT A GLANCE
Challenges
- Vendor access to sensitive data was blocked by legal and regulatory issues
- AI / ML projects were stalled as a result
Results
- 95% reduction in time to access sensitive data
- No performance loss
- Many new projects unlocked
"Subsalt eliminates the regulatory constraints from HIPAA on using my data, letting my team and vendors move much more quickly without exposing us or our patients to more risk."
Nathan Patrick Taylor
Chief Information Officer
