PRISM's 7-agent System Cuts NET Treatment Rejection Rate From 18% To 8%

PRISM's 7-agent System Cuts NET Treatment Rejection Rate From 18% To 8%

The Common Question

Can AI agents improve tumor board treatment recommendation quality in oncology?

The Lesson in This Case

PRISM, a 7-agent AI system tested across five European neuroendocrine tumor centers, cut the outright rejection rate of its treatment recommendations from 18% to 8% when evaluated by international experts in a blinded comparison, while pushing treatment completeness from roughly 70% to roughly 90%. The system also produced more than twice as many citations per recommendation as a standard LLM and generated zero fabricated references, pointing to structured multi-agent architecture as the driver of quality gains. Senior leaders evaluating AI for clinical decision support should note that PRISM's results rest on 15 offline cases, a sample small enough that a larger prospective trial is the essential next step before drawing budget conclusions.

Originally reported by AuntMinnieEurope. Read the full story here.