Abstract
OBJECTIVE: We evaluated the clinical performance of Roche Digital Pathology Dx, a whole slide imaging (WSI) system, in 2 studies according to US Food and Drug Administration (FDA) and Digital Pathology Association criteria. METHODS: Precision was measured by pathologists identifying 23 histopathology features; accuracy was assessed by comparing diagnoses from 2047 clinical cases with those from manual microscopy, with exploratory analyses including subgroup-specific diagnostic discrepancy rates. RESULTS: Both studies met all predetermined primary endpoints. Precision between systems/sites was 89.3%; between days, 90.3%; and between readers, 90.1% (lower bound of 95% CI for each, ≥85%). The difference in accuracy between digital reads (DRs) and manual microscopy reads (MRs) vs reference sign-out diagnosis (SD), DRs - MRs, was -0.61% (lower bound of 95% CI, -1.59%), which was greater than the lower bound acceptance criterion (-4%). Mean case reading times were similar: 2.33 minutes (DRs) and 2.34 minutes (MRs). Review of breast, lung, bladder, kidney, and stomach case diagnoses did not identify DR modality-specific root causes for major diagnostic disagreements. Higher than expected disagreements in both modalities were traced to COVID-19 pandemic-related resource constraints, leading to challenging case adjudications and higher disagreement rates for longer SDs. Direct DR/MR adjudication supported this hypothesis, resulting in an intermodality disagreement rate of 4.77%; using SD as a "tiebreaker" reduced the overall DR disagreement rate to 2.97%. CONCLUSIONS: Roche Digital Pathology Dx is noninferior to manual microscopy for primary diagnosis in surgical pathology, with performance results similar to 5 distinct FDA-cleared WSI systems using different scanners.