ncbi-hackathons/cteligible

Use machine learning to find patterns of similar eligibility protocol criteria for clinical trials

CTEligible

A Tool for Automating Similar Eligibility Protocol Criteria Writing for Clinical Trials Using Machine Learning

CTEligible is a tool for clinicians to automatically generate or write clinical trial protocols. The tool therefore solves the challenges and problematic or error prone nature of the manual process of writing clinical trial protocols.

CTEligible requires very minimal computer resources and can quickly generate clinical trial protocols.

CTEligible enables clinicians to do the following:
(1) Identify eligibility criteria that matches user inputs and makes recommendations,
(2) Reduces Drug Pricing, as this is one of the top 4 HHS Priorities

What's the project background?

Less than 5% of patients eligible for clinical trials register, resulting in higher drug development costs, drugs based on limited/overrepresented populations, and longer wait times. Making clinical trial search more effective can potentially decrease barriers to enrollment. The writing of clinical trial protocols -- the registration information describing a trial, its goals, the plan for carrying out the study, and who is eligible – historically has been written as free text independently for each trial. As a result, this text is not easily read by a computer. Searching across trials or linking data (like linking patients to eligible trials, or trials to eligible patients) becomes extremely problematic and error prone. One can make clinical trial searching and matching more successful by helping machines better understand the information that describes who is eligible for a clinical trial.

The goal of this project is to use machine learning techniques to find patterns of similar eligibility protocol criteria for clinical trials and guide the development of data-base clinical trial protocols for patient-to-trial matching.

CTEligible Intro

CTEligible Workflow

CTEligible Modular Workflow

Clustering Results

User Interface (UI) Challenges

A proposed framework at the 12th NIH Research Festival Collaborative Data Science and Machine Learning Hackathon on September 10 - 12, 2018

by Kelechi Mezu (ninabina921), Christopher Lavender (lavenderca), Julio Marco Pineda (juliomarcopineda), Max Serpe (maxSerpe), and Jia-Ling Lin (linjialing)

led by Justin Koufopoulos (jkoufopoulos) and Gil Alterovitz (gilusa)