H&E-based MSI/MMR testing with AI in colorectal cancer: a multi-centred blinded evaluation.

Mismatch repair (MMR) deficiency occurs in 10-20% of colorectal cancer (CRC) cases, leading to microsatellite instability (MSI). Although MSI/MMR testing is critical for CRC management, high costs and long turnaround times limit testing rates and clinical utility, highlighting the need for more accessible, cost-effective alternatives. PANProfiler Colorectal (PPC) is an artificial intelligence (AI)-based biomarker test that determines MSI/MMR status directly from haematoxylin and eosin (H&E)-stained slides. We conducted a blinded, multi-centred validation to assess PPC's performance against standard testing. The study included 3,576 whole slide images from 1,243 CRC patients across three United Kingdom institutions. PPC produced definitive results for 86.55% of slides, achieving an overall agreement of 93.83%, positive agreement of 92.54%, and negative agreement of 94.02%. PPC accurately determined MSI/MMR status from routine H&E slides, offering a rapid, scalable alternative to conventional diagnostic methods.