AI-driven resolution of uncertain histopathology in cases initially misdiagnosed as triple-negative breast cancer

Tumors presumed to be metastatic triple-negative breast cancer (TNBC) may occasionally represent misdiagnoses of primary or metastatic tumors from other origins—particularly lung—due to overlapping clinical characteristics and lack of organ-specific morphological features. While immunohistochemical markers can resolve these discrepancies, they are often underutilized without prior clinical suspicion, leading to significant treatment delays. To address this, we employed GPSai, a tissue-of-origin artificial intelligence (AI) model integrated into routine molecular profiling at Caris Life Sciences, to systematically identify differential diagnoses among 2423 presumed TNBC cases. By integrating AI results with clinical and molecular evidence, we identified a misdiagnosis rate of 3.0%. While over half of these cases were reclassified as non-small cell lung cancer, the misdiagnosis cohort also included various other tumor types. These diagnostic shifts profoundly impact staging, prognosis, and therapeutic selection, highlighting the necessity of an unbiased, systematic approach to identifying misdiagnoses among presumed TNBC.