Background: Among recently developed blood-based MCED tests, the ability to determine the location of tumors is pivotal to guiding appropriate treatment. We systematically reviewed and statistically examined the accuracy of tumor of origin predictions among blood-based MCED tests. Methods: Original articles were searched from Pubmed, Cochrane, and Embase for blood-based screening tests, multiple cancer types, and asymptomatic human subjects. We excluded studies with small samples (n < 30), non-screening, and non-blood-based tests. For cfDNA-based assays, measurements of diagnostic accuracy were pooled for meta-analysis. Results: Of 1,074 records identified and screened, five case-control studies and one cohort study that used cfDNA-based diagnostic tests were analyzed. Accuracy of tissue-of-origin (TOO) prediction for 3,895 cancer samples across cancer types was 0.79 (95% CI 0.66 - 0.90). Among six cancer types, colorectal cancers had the highest accuracy and liver & bile duct cancers had the lowest, although the difference was statistically insignificant (0.89 (95% CI 0.79-0.97) vs. 0.68 (95% CI 0.40-0.90)). Additionally, cases were most frequently misclassified as colorectal cancer (Table 1). The information for localizing TOO was derived from methylation patterns of cfDNA in four studies, fragmentation profiles of cfDNA in another study, and combination of mutations in cfDNA and protein markers in the last study. Conclusions: Our results demonstrate that the primary site of cancers was accurately discerned in 79% of cases by MCED tests. However, performance varies across cancer types. Further research on performance based on cancer stages and in combination with other molecular profiling is warranted.

Abbreviations: CI = confidence interval; GB = gallbladder, Upper GI = stomach and esophagus.