Building Patient Journeys from Multi-Modal Healthcare Data Using Medical Language Models

Data integration has been an enormous challenge in healthcare for decades

Issues of data standardization, data quality, legacy formats, unstructured data, and semantic inconsistencies have made it hard to answer basic questions about how a hospital operates or what should be done next for a patient. Recent advances in Healthcare AI combine to transform this age-old problem – enabling you to automatically ingest large volumes of raw, multi-format, multi-modal, untrusted medical data into coherent longitudinal patient stories in an industry-standard format.

This webinar presents an integrated solution in action that uses John Snow Labs’ state-of-the-art Medical Language Models, healthcare-specific data preparation pipelines, and Text-to-OMOP question answering models running on Databricks’ secure, scalable, and compute-optimized AI platform. The solution takes in multi-modal data – structured (tabular), semi-structured (FHIR resources), and unstructured (free-text) – and generates an OMOP/OHDSI standard data model that: 

• Builds a unified view of each patient over time.
• Builds this unified patient view from multi-modal source data.
• Reasons at the patient level.

We’ll then show how the resulting patient data model can then be used for either “AI” (building patient cohorts with natural language queries) or for “BI” (dashboards for patient risk scoring and quality measures), all from the same source of truth, with full explainability and traceability.

PRESENTED BY:

Kayla Grieme
Sr. Solutions Architect - Academic Medical Centers
Databricks

Kayla Grieme is a Sr. Solutions Architect at Databricks, working with Academic Medical Centers to leverage data science and AI to drive innovation and improve patient outcomes through advanced analytics solutions.

Kate Weber
Senior Data Scientist
John Snow Labs

Kate Weber is a Senior Data Scientist at John Snow Labs who specializes in healthcare natural language processing and data standards. While completing her Ph.D. at the University of Michigan, she built algorithms to detect and classify evidence of substance use disorder in clinical notes, and pioneered approaches to using artifacts in the data annotation process to get the most out of precious labelled resources. Her background in technical infrastructure and data engineering helps her understand the scope of the challenge facing enterprise health informatics teams. On her own time, she races bicycles and maintains the technical infrastructure for her family's home-brewing and beekeeping adventures.