A large-scale, multi-centre validation study of an AI-empowered blood-based test for multi-cancer early detection

Abstract Cancer is a critical global health issue, especially in low- and middle-income countries (LMICs). In this study, we integrated four additional cohorts to assess the performance and robustness of an AI-empowered blood-based test (named OncoSeek) for multi-cancer early detection (MCED). It included a case-control cohort of symptomatic cancer patients, a prospective blinded study, and two retrospective cohorts conducted on two distinct platforms. Combining these with previously published one training and two validation cohorts, we evaluated OncoSeek’s performance in 15,122 participants (3029 cancer patients and 12,093 non-cancer individuals) from seven centres in three countries, using four platforms and two sample types. OncoSeek showed adequate performance for MCED with an area under the curve (AUC) of 0.829, 58.4% sensitivity, 92.0% specificity, and overall accuracy of 70.6% in tissue of origin (TOO) prediction for the true positives. The test could detect 14 common cancer types, accounting for 72% of global cancer deaths, with sensitivities ranging from 38.9 to 83.3%. Additionally, the symptomatic cohort exhibited a high sensitivity of 73.1% at 90.6% specificity, indicating OncoSeek’s potential for cancer early diagnosis. These findings underscore OncoSeek’s consistent performances across diverse populations, platforms, and sample types, offering affordable and accessible multi-cancer early detection, especially for LMICs.