Is your infrastructure standing in the way of better outcomes for your organization and the customers you serve?
Baylor College of Medicine, a large academic medical hospital, was facing a challenge: massive amounts of its data were locked away in unstructured, free-form text medical notes. Without costly manual extraction, Baylor’s researchers, clinicians and administration personnel were prevented from fully utilizing this invaluable source of information. A team from Pariveda designed a cutting-edge natural processing pilot to extract the data.
A data pipeline to deeper insight and analysis
The team built a data pipeline using a combination of on-premises and cloud-based technology to accomplish the following objectives:
· De-identify patient records by removing all 18 categories of personal health information, or PHI, to comply with HIPAA regulations;
· Use the cutting-edge cloud-based AWS Comprehend Medical Machine Learning/Artificial Intelligence algorithm to perform NLP on the notes to extract usable, structured data from the previously unstructured text and make it available to the hospital’s existing analytics toolchain; and
· Showcase results across the organization to increase excitement around ML technologies and determine value-drive concrete use cases to explore.
Data-driven system improves both operations and outcomes
When fully implemented, the platform benefits the Baylor organization and its patients by:
· Exposing missed revenue opportunities by finding over or under-coding billing claims, which leads to higher revenue and lower risks of fraudulent claims;
· Reducing the costs associated with using symptoms, signs, medication and medical history in cohort identification for clinical trials. In turn, this leads to finding more eligible participants in days rather than months, accelerated study timelines, increased time and funds available for analysis and, ultimately, higher quality studies translated into improved healthcare for patients;
· Increasing patients’ quality of care through the facilitation of iterative reduction and avoidance of side effects through data-driven quality efforts using information previously difficult to access; and
· Reducing for clinicians the time required to write referral notes or populate a charge screen by predicting diagnoses and billing codes from the progress/encounter note.