Using Generative AI to Extract SDOH

Problem Background

Recognizing the critical role of inclusive data and cultures when approaching healthcare challenges and reducing health inequities, two leading healthcare organizations collaborate on an approach to identify Social Determinants of Health (SDOH) in medical data. This new collaboration aims to co-validate technology approaches labelling SDOH factors such as socioeconomic status, housing, and transportation.

Proposed Solution

eXtracting Determinants of Health (XDOH) allows clinical staff to gain valuable insights and connect patient populations to needed services automatically. The XDOH model has been validated across multiple healthcare organizations to help reduce bias in the model, and explore shared commonalities between global healthcare systems. XDOH easily integrates with medical systems and provides an automated pipeline to allow for ingesting standardized FHIR data as well as other unstructured data. This data is then processed by the XDOH ingestion engine. This data is evaluated for SDOH and then determinations are pushed back to the originating health care system. The XDOH platform allows for continuous training through a data feedback loop as well as human-in-the-loop opportunities to evaluate and audit model performance.

Healthcare Innovation CIC Takes on Initial Prototyping

The Healthcare Innovation CIC worked closely with two leading medical centers in the US and Asia to develop an approach to prototype the XDOH concept. The US healthcare center explored how to utilize Generative AI to extract SDOH from unstructured medical notes, while the Asia healthcare center aimed to explore how structured data could be utilized to trigger insights into SDOH.

Storyboarding

The following represents a visual journey of the customer problem, solution discovery, and solution implementation.

Prototyping

The team decided to focus efforts on one primary output:

  1. A data generator utilizing generative AI
  2. A SDOH detector utilizing generative AI

Call to Action

Interested in integrating the concepts of this challenge into your production workloads? Contact us and we can discuss how AWS Cloud and the AWS Partner Network can help accelerate your goals.

DeliverableDescriptionLink
PRFAQFictional press release for the XDOHhttps://github.com/UC-CIC/SDOH-extraction-doc/blob/main/prfaq.pdf
StoryboardVisual journey of the concepthttps://github.com/UC-CIC/SDOH-extraction-doc/blob/main/storyboard.pdf
Conceptual ArchitectureConceptual cloud architecture diagram of how XDOH could be deployedhttps://github.com/UC-CIC/SDOH-extraction/blob/main/conceptual_arch.pdf
Prototype (Technical)Technical prototype utilizing Amazon Bedrockhttps://github.com/UC-CIC/SDOH-extraction