CTSuggest: Clinical Trial Suggestion of Baseline Features using LLM

Deployed version at: https://inciteprojects.idea.rpi.edu/ctsuggest/app/ctsuggest/

Introduction

CTSuggest is a user-friendly application that leverages Large Language Models to generate baseline features for clinical trial design. CTSuggest enables users to specify trial metadata, generate baseline feature suggestions with explanations, and evaluate performance using an "LLM-as-a-Judge" approach. Users can create new trials or use existing trial data from ClinicalTrials.gov. CTSuggest builds upon and improves the LLM prompts developed in the successful CTBench Benchmark (https://github.com/nafis-neehal/CTBench_LLM). Evaluations using reference baseline features taken from clinical trial publications show that CTSuggest provides suggestions using the state-of-the-art GPT-4o model and the local llama model of RPI. CTSuggest is an open-source app for baseline feature generation with explanations available to trial designers.

Setup & Deployment

Clone the repository

git clone https://github.rpi.edu/DataINCITE/CTSuggest.git

Check for R installation
```
R --version
```
Navigate to the project root and open it in your preferred code editor
Install required libraries
- From command line interface:
```
Rscript setup_dependencies.R
```
- From R terminal:
```
source('setup_dependencies.R')
```
Run the Shiny application
- From command line interface:
```
Rscript app.R
```
- From R terminal:
```
shiny::runApp('app.R')
```

Operation Instructions

Step 1: Specifiy Trial

Fill all specified fields by:
- Selecting a trial from the dropdown menu under Choose an NCT ID: to load it's features and changing them as desired; or
- Manually entering the fields to your trial specifications.

Step 2: Generate Descriptors

Click Generate to call an LLM to generate a list of baseline descriptors for the specified trial. The output will appear below.
Navigate to Options to modify the generation options.

Step 3: Evaluate

Click Run Evaluation to call an LLM to compare the list of generated baseline features to the baseline features actually measured in the trial.
Note: Evaluation is only available when features have been generated for an existing trial

Please see the wiki for more details.

DataINCITE/CTSuggest

About

Resources

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors 6

Uh oh!

Languages

Packages