Mendel reports positive results in AI cohort matching trial

Mendel, a clinical AI firm, has announced results from a trial showing its Neuro-Symbolic system outperforms current models like GPT-4 in sorting patient cohorts from electronic medical records (EMR). Conducted with Cornell University, titled ‘ACR: A Benchmark for Automatic Cohort Retrieval’, the study unveils integrating expert knowledge with language learning models (LLMs) in healthcare.

Mendel’s AI combines LLMs with a proprietary hypergraph reasoning engine to analyze patient cohorts and identify clinical trial candidates. This capability is crucial for forming effective trial cohorts. The trial evaluated 1,400 patient records.

In another study, Mendel collaborated with nurses to speed up oncology patient pre-screening for trials. Combined AI and human accuracy matched humans alone (78.7% vs. 76.7%), surpassing AI-alone (63.5%).

The findings underscore the potential of neuro-symbolic approaches to enhance ACR systems in healthcare by providing system control and reliable performance, addressing real-world needs effectively.