This is a demo of how to leverage Generative AI to perform cohort identification by matching patients to clinical trials.
Clinical Trials Gen AI workshop Part 1.ipynb contains a Jupyter notebook file that walks through step-by-step how to use Amazon Bedrock and Large Language Models (LLM) to solve a real-world problem described in the User persona below. Clinical Trials Gen AI workshop Part 2.ipynb is a continuation of Part 1, where we will move the use-case closer to production by focusing on evaluation logic, chaining LLMs and cost analysis. Clinical Trials Gen AI Part 3.ipynb located inside the Part 3 folder is a continuation of Part 1 where we leverage batch inferencing to perform clinical trial patient matching at scale in conjuction with an EHR system.
AnyCompany is a healthcare startup that provides insights based on patient history. They offer a solution that can be embedded into an EMR, so their customers include EMR vendors and healthcare providers. They offer their solution as a SaaS model and their infrastructure is cloud native and built leveraging serverless to provide elasticity and scalability. The Product team recently flagged the need for the company to offer Cohort Identification as a service and their first investment is to build a Clinical Trials matching feature. The business is attempting to evaluate if a clinical trial is suitable for a patient given a patient's longitudinal records.
As a data scientist recently hired by AnyCompany, you got pulled into a tiger team that has the goal of (i) validating if this project is feasible, (ii) building a working proof of concept.
If your assignment succeeds, this feature will be added to AnyCompany's next big release, which is coming out soon.
The clock is ticking, so let's start building!
Using.Generative.AI.to.Match.Patients.with.Clinical.Trials.Walkthrough.480p.mov
The notebook requires multiple resources to be enabled or created.
- Ensure access to Anthropic models.
- Ensure proper IAM permission to invoke models and subscribe to 3rd party models.
- A IAM role which may need to be created is required to use Bedrock Batch Inferencing in Part 3. Details are explained in the jupyter notebook where required.
- Creation of Amazon Healthlake with Synthea data enabled and FHIR API permission enabled - (Optional for part 1 and 2, highly encouraged for part 3)
- Notebook Environment requires Python version >= 3.10.0
- Compute resource similar to Sagemaker Studio's ml.t3.medium instance 2vCPU and 4Gib Memory
- Refer to requirements.txt for package requirements
See CONTRIBUTING for more information.
This library is licensed under the MIT-0 License. See the LICENSE file.
Jason Walonoski, Mark Kramer, Joseph Nichols, Andre Quina, Chris Moesel, Dylan Hall, Carlton Duffett, Kudakwashe Dube, Thomas Gallagher, Scott McLachlan, Synthea: An approach, method, and software mechanism for generating synthetic patients and the synthetic electronic health care record, Journal of the American Medical Informatics Association, Volume 25, Issue 3, March 2018, Pages 230–238, https://doi.org/10.1093/jamia/ocx079