Case Studies

How an International Wellness Giant Modernized its Clinical Trial Screening Processes

Azure clinical trials data in healthcare snowflake cortex Streamlit

faster patient pre-screening

x 0 x

efficiency gains in 1st tier pre-screen determinations

0 %

build speed vs. industry standard

0 x

Designed and implemented Snowflake architecture capable of 3rd party data integration.

Vetted external vendors and advised on mission-critical capabilities like de-identification, OMOP modeling, and patient benchmarking.

Leveraged Snowflake NLP and LLM functionality to design and implement a Clinical Trial Waiting Room that streamlines pre-screening processes.

Challenge

The client was the wellness division of a large international corporate group, which covers businesses spanning the healthcare, pharmaceutical, and hospitality industries. At the beginning of the engagement, the customer’s teams were struggling to measure and benchmark patient outcomes due to struggles with using unstructured data and limited analytical, benchmarking, and ML development capabilities.

The customer also faced challenges in its clinical research operations, where its teams were encumbered by a heavily manual screening process to identify clinical trial participants among their current patients. To ameliorate these bottlenecks, the customer was on the lookout for tools to automate those processes and foster collaboration with sponsors on study creation.

Solution

To help the client’s internal teams unlock more robust analytical and ML capabilities, Hakkoda designed and implemented end-to-end architecture based on Snowflake and Azure. This implementation included vetting 3rd party data integration and build-or-buy testing with vendors across the modern technology stack for capabilities ranging from de-identification, to OMOP modeling, to patient benchmarks, to LLM interaction hosted in a user-friendly streamlit app.

Hakkoda also developed a comprehensive Power BI dashboard that seamlessly integrates and displays all patient information, from SDOH to clinical, providing physicians and operations teams with comprehensive insight into each patient’s medical history and a 360-degree view of custom patient populations.

The resulting data stack, which leveraged a host of Snowflake features including Snowpipe and Cortex, enabled data flow and NLP/LLM capabilities that would empower physicians and operation leads with immediate education and pre-screening for all active patients, including flagging them for trial feasibility. To accomplish this, the Hakkoda team utilized data science techniques like Named Entity Recognition, Fuzzy Matching, and the Sorensen Dice Index to enhance and strengthen the NLP portion of the model.

Building on these NLP and LLM capabilities in an explainable and accurate manner, Hakkoda worked with the customer to develop a data product domain, the Clinical Trial Waiting Room, to align all operational and data activities around the patient. This Clinical Trial Waiting Room, once implemented, empowers internal talent with the ability to see, segment, and serve underserved populations in all of their clinical trial initiatives.

To understand patient survival rates, the Kaplan-Meier method was instituted, providing a deeper understanding of trial and treatment effectiveness. This solution, meanwhile, leveraged Azure Dev Ops to streamline and automate the whole process. When a new patient is added to a table, the MLOps process automatically triggers a model run on that patient, providing the end user with a model output for which trials that patient may be a good fit.