This is really a spinoff through a program called CancerX, where we’re co-developing this AI platform with another company called LendAI, a startup company. And during this co-development process, really the first phase of this is to validate that platform using real-world data from our own patient records, as well as real cancer trial protocols. And this has really been one of the gaps in prior literature that have mostly focused on synthetic data sets...
This is really a spinoff through a program called CancerX, where we’re co-developing this AI platform with another company called LendAI, a startup company. And during this co-development process, really the first phase of this is to validate that platform using real-world data from our own patient records, as well as real cancer trial protocols. And this has really been one of the gaps in prior literature that have mostly focused on synthetic data sets. So we took an opportunity to take thousands of patients within our own health system, develop the model, train the model, use machine learning plus large language models, kind of a hybrid architecture to further train that model and then test it in a separate independent test set of patients. So the test set of patients focused on individuals with breast cancer, lung cancer, and pancreatic cancer. So we were matching these patients to retrospective clinical trial protocols. In that study, we identified that there was actually 94% agreement or concordance between a clinical research coordinator reviewing the data versus the AI platform, so performing really just as well as the human review. So at different probability thresholds, we then determined that the platform can have upwards of 99% sensitivity and positive predictive performance really upwards of, again, 90% or so, meaning that every nine out of every 10 patients or so that the AI platform says the individual is eligible for the clinical trial, they truly end up being eligible for that particular clinical trial.
This transcript is AI-generated. While we strive for accuracy, please verify this copy with the video.
