The Potential of AI in Data Collection

ACT: Is there any potential for the use of artificial intelligence (AI) in the data collection space?

Gliklich: I love that question because we literally power our data automation and data collection with AI in multiple ways. It starts with automating data mapping and quality assurance. When we bring data in from a medical record system or a laboratory information management system, we need to bring that into a common data model by leveraging generative AI, we can do that much faster and then verify it. Where we see its greatest benefits today are in collecting and transforming unstructured data, like clinical notes, imaging reports, pathology reports, procedure reports, into structured variables, so capturing MRI lesions or symptoms, those are areas that we’re using today at scale. We still have to go and validate it with extractors to demonstrate, at least in a sample, that it’s doing the right thing. That’s one area. Another area that we find really exciting is the concept of using clinical narratives for generating estimated or surrogate endpoints, so estimated endpoints, or surrogate endpoints, as you know, can speed a trial or make more patients useful in a trial, and we can derive those from the medical record patterns that we see using AI or from an estimation.

We’ll take a clinical note, and we will use patients where or visits where a measurement has been recorded, and those where it hasn’t been recorded, and the system will actually read the clinical note, it’s almost magical, and estimate an endpoint from that. We validate many of those and I was using some examples from lupus, the SLEDAI as an example, because it’s used in clinical trials, but it’s only used in like 5% of practice in the clinical world; 5% of encounters, so it’s incredibly valuable to be able to do that. Those are some of the areas where we see value and data collection.