Overview

Identifying cohorts of patients is a fundamental step in many healthcare applications. ScienceIO supports patient cohort selection with our healthcare AI solution. Identify patients based on key information in their unstructured data: diagnoses, vitals, test results, clinical documentation, and more. Speed up patient screening, clinical trial recruitment, and cohort identification for real-world studies. Our comprehensive AI platform can label medical documents with over 9M healthcare concepts.

Highlights

• Finding patients in electronic health record systems is a key step in many workflows, from trial recruitment to real-world data collection

• Patients with complex attributes are not easily findable and require manual review

• A single patient can have hundreds of events and documents, making manual review impractical

• AI-enriched patient records can make patients more easily findable for multiple applications and workflows

Cohort identification for clinical trial recruitment

Identifying patients that match a clinical trial’s inclusion criteria from EHR data can augment the recruitment funnel or find rarer patient populations. Often, recruitment criteria are complex and not searchable in the EHR. For example, there is no explicit diagnostic or procedure code unique to patients with triple-negative breast cancer. A search for breast cancer codes in an EHR may yield thousands of results that require manual review to determine the specific cancer subtype. The necessary information is encoded in unstructured records, such as a clinical diagnosis written in plain text or supporter biomarker reports in a PDF attachment.

In contrast, ScienceIO’s AI platform can make patients searchable by more complex criteria. Our AI identifies the key information on a patient from those unstructured records, which are then more easily searchable.

Case study: Rapidly identify complex cohorts on the Mayo Clinic Platform

Recently, we demonstrated the need for NLP tools to identify patient cohorts with high accuracy and speed on the Mayo Clinic Platform, which has millions of de-identified patient records which can be securely accessed. From a single diagnostic code, we identified more than 100k breast cancer patients. We estimated that a manual review of all these cases to confirm triple-negative breast cancer diagnoses would take 100+ work years. This is exacerbated by the fact that even a single patient can have hundreds of notes and documents. Instead, we indexed each patient’s clinical notes which revealed 55 confirmed cases in less than 1 day.

AI-enriched patient data can be used for multiple applications. In this example, the same dataset can be used to find patients with other subtypes of breast cancer or patients with specific biomarkers in pan-tumor settings.